MIT Theses

A design of a low-pressure steam turbine

Sat, 01 Jan 1910 00:00:00 GMT

A design of a low-pressure steam turbine Jones, Bradley. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1910

Computational study of static replication for barrier options

Wed, 01 Jan 1997 00:00:00 GMT

Computational study of static replication for barrier options Sun, Hai Po. Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 1997; Includes bibliographical references (leaves 75-76).

Modeling, control and experimentation of a two dimensional linear motor

Wed, 01 Jan 1997 00:00:00 GMT

Modeling, control and experimentation of a two dimensional linear motor Castañeda Vega, José Israel. Thesis: M.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1997; Includes bibliographical references (leaf 118).

A study of anode dimensions in mercury-vapour thermionic rectifiers

Fri, 01 Jan 1932 00:00:00 GMT

A study of anode dimensions in mercury-vapour thermionic rectifiers Fussell, Lewis. Thesis: M.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1932; Includes bibliographical references (leaf 50).

Cloud-chamber study of cosmic ray showers in lead plates

Sat, 01 Jan 1938 00:00:00 GMT

Cloud-chamber study of cosmic ray showers in lead plates Fussell, Lewis. Thesis: Sc. D., Massachusetts Institute of Technology, Department of Electrical Engineering, 1938; Includes bibliographical references (leaves [113]-[118]).

Development of a high-speed light source suitable for photoelastic studies

Wed, 01 Jan 1941 00:00:00 GMT

Development of a high-speed light source suitable for photoelastic studies Wyle, Frank S. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1941; Includes bibliographical references (leaf 25).

Boiling and spreading rates of instantaneous liquid methane spills on water

Fri, 01 Jan 1982 00:00:00 GMT

Boiling and spreading rates of instantaneous liquid methane spills on water Chatlos, David Joseph. Thesis: M.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1982; Supervised by Robert C. Reid.; Includes bibliographical references (leaves 86-88).

Manipulation and measurement of charge transfer kinetics at chemically modified electrodes

Thu, 01 Jan 1981 00:00:00 GMT

Manipulation and measurement of charge transfer kinetics at chemically modified electrodes Lewis, Nathan S. (Nathan Saul) Thesis: Ph. D., Massachusetts Institute of Technology, Department of Chemistry, 1981; Includes bibliographical references.

Application of Fourier transform spectroscopy to the absolute determination of the chemical shift of protons.

Wed, 01 Jan 1975 00:00:00 GMT

Application of Fourier transform spectroscopy to the absolute determination of the chemical shift of protons. Wright, Francine Elaine. Thesis: M.S., Massachusetts Institute of Technology, Department of Physics, 1975; Vita.; Bibliography: leaves 65-66.

Double valves

Mon, 01 Jan 1877 00:00:00 GMT

Double valves Faunce, Linus. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1877

Higher Siegel-Weil formulae over function fields

Sun, 01 Feb 2026 00:00:00 GMT

Higher Siegel-Weil formulae over function fields Mkrtchyan, Mikayel In their seminal work, Feng-Yun-Zhang introduced function field analogues of Kudla-Rapoport cycles for moduli spaces of unitary shtukas, and initiated the study of their intersection theory. They proved a higher Siegel-Weil formula in the case of non-degenerate Fourier coefficients, relating the degrees of these cycles to higher derivatives of Siegel-Eisenstein series. In this thesis, we generalize their result in two directions: we 1) prove a higher Siegel-Weil formula for unitary groups for corank-1 degenerate coefficients, and 2) introduce analogous cycles on moduli spaces of symplectic shtukas, and prove a higher Siegel-Weil formula for such cycles in the non-degenerate case, relating their degrees to derivatives of Siegel-Eisenstein series on split orthogonal groups.

A Diverse Array of Synthetic Strategies for Phosphorus Group Transfer Chemistry: From Phosphinidenes to Phosphates

Sun, 01 Feb 2026 00:00:00 GMT

A Diverse Array of Synthetic Strategies for Phosphorus Group Transfer Chemistry: From Phosphinidenes to Phosphates Xin, Tiansi This thesis compiles the published scientific contributions of Tiansi Xin. Chapter 1 consists of a brief collection of eulogies from friends and colleagues, reflecting on his life and time at the Massachusetts Institute of Technology. The subsequent chapters describe the development of novel synthetic methods for the transfer of phosphorus-containing moieties, specifically metaphosphates and phosphinidenes. The work presented here has significant implications for both the fundamental understanding and practical advancement of synthetic inorganic and organic chemistry. Chapters 2 and 3 address the sustainable production and processing of phosphoruscontaining chemicals, focusing on mechanochemical methods to synthesize reduced phosphorus species while circumventing the need to access hazardous white phosphorus as an intermediate. In particular, Chapter 2 describes a solvent-free mechanochemical approach to producing phosphite (HPO₃²⁻) via hydride-mediated reduction of condensed phosphates. Using potassium hydride, a range of inorganic phosphate sources—including pyrophosphate, triphosphate, trimetaphosphate, fluorophosphate, and polyphosphate—were converted to phosphite in moderate to high yields. Mechanistic studies identified overreduction pathways leading to hypophosphite and other low-oxidation P-species. Chapter 3 similarly applies this mechanochemical approach to phosphorus–carbon bond formation, reporting the phosphorylation of acetylides with condensed phosphates to afford phosphonates. Biogenic polyphosphates were also shown to be viable precursors, a proof-of-concept to closing the modern phosphorus cycle using recycled inputs. These results demonstrate the possibility of accessing organophosphorus chemicals directly from condensed phosphates and may offer an opportunity toward a “greener” phosphorus industry. Chapters 4 and 5 shift focus to phosphinidene transfer chemistry and the synthesis of novel phosphorus-containing heterocycles. This expands on previously published studies from the Cummins group on the chemistry of dibenzo-7-phosphanorbornadiene “RPA” reagents. Chapter 4 reports the preparation and structural characterization of iron–phosphido complexes relevant to phosphinidene group transfer catalysis and describes the development of an improved catalytic system based on a simple diiron precursor (Fp₂), enabling efficient synthesis of phosphiranes from electron-deficient alkenes. The mechanism was thoroughly experimentally and computationally interrogated. Chapter 5 describes the novel synthesis of free, uncomplexed phosphet-2-ones via phosphinidene transfer to cyclopropenones, with experimental and theoretical studies supporting a mechanism involving ketene-derived intermediates and transformations to additional phosphorus heterocycles through subsequent reactions.

Geology of Deception Gulch and the Verde Central mine

Mon, 01 Jan 1923 00:00:00 GMT

Geology of Deception Gulch and the Verde Central mine Benedict, P. C. (Platt Carrico), 1900-1969. Thesis: M.S., Massachusetts Institute of Technology, Department of Geology and Geophysics, 1923

Structural geology of Eastern Massachusetts

Mon, 01 Jan 1934 00:00:00 GMT

Structural geology of Eastern Massachusetts Ilsley, Ralph, 1896- Thesis: Sc. D., Massachusetts Institute of Technology, Department of Geology, 1934; Vita.

The design of a hydraulic draft gear for railway passenger cars

Sun, 01 Jan 1922 00:00:00 GMT

The design of a hydraulic draft gear for railway passenger cars Pearson, Harry L.; McGrady, Charles T. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1922; Includes bibliographical references.

A study of "Chu-Ma" as a textile fiber

Tue, 01 Jan 1924 00:00:00 GMT

A study of "Chu-Ma" as a textile fiber Chou, Cheng Yu, 1901-; Hsueh, Tsu Kang. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1924

The transportation decision making process in metropolitan Boston

Sun, 01 Jan 1967 00:00:00 GMT

The transportation decision making process in metropolitan Boston Zinner, Richard Mark. Thesis: B.S., Massachusetts Institute of Technology, Department of Political Science, 1967; One unnumbered page inserted.; Bibliography: leaf 74.

Oscillographic presentation of impedances on the reflection-coefficient plane

Sat, 01 Jan 1949 00:00:00 GMT

Oscillographic presentation of impedances on the reflection-coefficient plane Eckhart, Myron.; Fowler, Earl Bealle. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1949

Radiation transfer in massive binary x-ray systems

Tue, 01 Jan 1991 00:00:00 GMT

Radiation transfer in massive binary x-ray systems Lewis, Wayne Lloyd. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Physics, 1991; Includes bibliographical references (leaves 167-173).

Laser induced photoionization of helium

Tue, 01 Jan 1980 00:00:00 GMT

Laser induced photoionization of helium Lewis, Wayne Lloyd. Thesis: B.S., Massachusetts Institute of Technology, Department of Physics, 1980; Includes bibliographical references.

Nonlinear elastic analysis of reinforced concrete structures by the finite element method

Mon, 01 Jan 1979 00:00:00 GMT

Nonlinear elastic analysis of reinforced concrete structures by the finite element method Tulga, Said Şahin. Thesis: M.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1979; Includes bibliographical references.

Subcontractor bidding strategy

Wed, 01 Jan 1975 00:00:00 GMT

Subcontractor bidding strategy Gilbane, Thomas Freeman. Thesis: M.S., Massachusetts Institute of Technology, Department of Civil Engineering, 1975; Bibliography: leaves 104-105.

Petrography and geology of the Shoshone mining region in northwestern Wyoming

Sun, 01 Jan 1922 00:00:00 GMT

Petrography and geology of the Shoshone mining region in northwestern Wyoming Benedict, P. C. (Platt Carrico), 1900-1969. Thesis: B.S., Massachusetts Institute of Technology, Department of Geology, 1922

Mechanisms of Interaction Between Hydraulic and Natural Fractures in Shale Rocks

Mon, 01 Sep 2025 00:00:00 GMT

Mechanisms of Interaction Between Hydraulic and Natural Fractures in Shale Rocks Arzuaga García, Ignacio Martín Understanding the interaction between hydraulically induced fractures and pre-existing natural fractures in geologic formations is key for optimizing subsurface energy systems that rely on fluid injection into fractured rocks. These include Enhanced Geothermal Systems (EGS), CO₂ sequestration, hydrogen storage in depleted reservoirs, unconventional oil and gas development in shale formations, and nuclear waste disposal, among others. In all these applications, controlling fracture propagation and interaction is essential for ensuring operational efficiency, safety, and long-term integrity of the system. This thesis presents a comprehensive experimental and theoretical investigation of hydraulic fracture (HF) interactions with natural fractures (NFs), using Opalinus Clayshale as a representative anisotropic material. The experimental work involved a series of hydraulic fracturing tests on Opalinus Clayshale specimens under controlled quasi-true-triaxial stress conditions, comparing normal and dried states. Novel monitoring techniques, including high-resolution imaging, high-speed video, acoustic emissions (AE), and pressure tracking, were employed to capture the fracturing process in real-time. Three dominant interaction modes (Crossing, Arrest, and Opening) were systematically characterized and linked to key parameters, including stress ratio, fracture geometry, and injection rates. A critical stress ratio (σ₁/σ₃) of approximately 20 was identified as the threshold for achieving fracture crossing under our experimental conditions: cohesionless, “open” natural fractures, with a low viscosity injection fluid, in a toughness-dominated regime. In dried specimens, high flaw pressurization rates were necessary to overcome matrix fluid loss and achieve crossing. To complement and interpret the experimental results, existing theoretical models were reviewed and implemented. Furthermore, a simplified version of the OpenT model (Chuprakov et al., 2014) was developed and applied for Opalinus Clayshale, incorporating stress, energy, friction, and permeability effects. By integrating laboratory results with theoretical frameworks, this thesis offers an integral approach to predictive understanding of fracture propagation in naturally fractured rocks, stating that not only the characteristics of the discontinuity or the far-field stresses involved in the process are important in determining the mechanism of interaction, but also the dynamic energy balance at the fracture tip, which is influenced by injection rate, fluid viscosity, and discontinuity properties. Overall, this thesis bridges the gap between laboratory experiments and theoretical models, advancing a more comprehensive understanding of fracture propagation in naturally fractured media. The findings highlight the importance of considering both mechanical and hydraulic parameters, particularly in low-viscosity, toughness-dominated regimes, for accurately predicting fracture behavior.

Satellite Drag and Sustainable Space Operations in a Dynamic Thermosphere

Mon, 01 Sep 2025 00:00:00 GMT

Satellite Drag and Sustainable Space Operations in a Dynamic Thermosphere Parker, William E. Earth’s orbit has become increasingly congested and contested in recent years. The surge in launched payloads, combined with satellite failures, explosions, and collisions, has contributed to a large and growing population of orbital debris objects that can remain in orbit for decades, centuries, or longer. Meanwhile, decreasing launch costs and maturing satellite technology have created conditions favorable for rapid commercialization across orbital regimes, especially in low Earth orbit (LEO). Today, a small number of commercial entities operate the large majority of the world’s active satellites as part of proliferated LEO constellations. Sustaining productive activity in an increasingly crowded orbital environment has made satellite conjunction assessment and collision avoidance essential for safe operations. These efforts require not just accurate trajectory predictions, but also credible estimates of uncertainty. In LEO, variability in atmospheric drag is by far the dominant source of propagation error, often leading to deviations of several kilometers per day due to unpredictable solar and geomagnetic activity. Even over short timescales, trajectory prediction is challenging because existing forecasts exhibit limited predictive skill. Although forecast errors are often non-Gaussian and heteroscedastic, operational products are generally presented as deterministic, and atmospheric models rarely provide rigorous uncertainty characterization. This work introduces a new approach for probabilistic satellite drag modeling based on historical correlations between space weather drivers and satellite dynamics. Unlike traditional methods, it models satellite behavior directly without reconstructing thermospheric mass density or requiring detailed knowledge of satellite properties such as the ballistic coefficient. This end-to-end strategy offers substantial computational and operational advantages for many space domain awareness tasks. Capturing both trajectory predictions and their associated uncertainty is critical for enabling informed collision avoidance decisions, particularly during geomagnetic storms when current infrastructure frequently fails. Because the orbital lifetime of debris objects can exceed hundreds of years, population dynamics in space critically depend on long-term variability in the composition of Earth’s thermosphere. Rising concentrations of carbon dioxide and other greenhouse gases have caused warming in the troposphere but cooling and contraction in the upper atmosphere. This contraction decreases atmospheric density in LEO, reducing drag and extending the orbital lifetime of debris objects. Longer-lived debris populations pose a persistent collision hazard for all active satellites as long as they remain in orbit. Even natural events, such as a prolonged grand solar minimum, could further reduce thermospheric density and contribute to longer debris lifetime in LEO. With little ability to predict such an event, it is necessary to understand the potential consequences and to identify strategies that enable the continued safe and productive use of LEO. This work models the impact of such long-term environmental changes on limits for sustainable satellite deployments. LEO is a finite respource increasingly at risk of overexploitation. Conserving it and sharing it fairly requires that we first understand its fundamental capacity and our current occupation of that capacity. Some metrics have been proposed to measure the satellite carrying capacity of Earth’s orbit, but none have previously accounted for the potential influence of a changing space climate. This work develops new methods for defining carrying capacity as a common currency, enabling clear constraint-driven thresholds on activity and a better understanding of how existing and proposed missions consume available capacity. These new metrics provide insight into how environmental variability may affect the long-term sustainability of operations in LEO. Respecting and understanding this influence that the natural environment has on our collective ability to operate spacecraft in LEO is critical to preventing the overexploitation of this regime and protecting it for future generations.

Evaluating Large Language Models as Circuit Design Assistants

Sun, 01 Sep 2024 00:00:00 GMT

Evaluating Large Language Models as Circuit Design Assistants Cox, Matthew J. Large language models (LLMs) have exploded in capability in recent years. Previous attempts at AI systems for circuit design have had limited proficiency and been restricted in problem scope. LLMs, with their breadth of knowledge and reasoning ability, are a promising technology for a much more general-purpose circuit design assistant. We developed a dataset of electrical engineering problems and solutions with which to test an LLM-based system, since no such publicly available dataset exists to our knowledge; unmodified GPT-4 was able to solve 42% of the problems. We did a preliminary comparison of several knowledge bases to use for RAG knowledge injection, finding that a small, curated set of resources performed better than a larger, less-focused set of resources, though there were confounding factors which may have skewed the result. While this work is a start, significant future work is needed to continue developing an LLM-based circuit design assistant.

Increasing Program Code Coverage Using Fuzzing and Targeted Branch Exploration

Mon, 01 Sep 2025 00:00:00 GMT

Increasing Program Code Coverage Using Fuzzing and Targeted Branch Exploration Nguyen, Gary Code coverage is a longstanding metric for evaluating how thoroughly a program has been tested. Achieving high coverage remains a priority goal for quality assurance and software stability. Exhaustive enumeration of possible input paths to every code region is desirable in theory but computationally infeasible in practice, especially in large-scale codebases. Fuzzing is a widely used technique for input generation and is effective at exploring smaller programs but often struggles with more complex conditional logic and nested modules. Concolic execution, which exhaustively explores paths using constraint solving, can work effectively with complex conditional logic but suffers from path explosion. Targeted branch exploration is a similar approach for input generation but sidesteps the path explosion problem by focusing more on specific constraint paths of interest. In this thesis, I introduce a hybrid system that combines fuzzing and targeted branch exploration with the goal of improving code coverage by leveraging the complementary strengths of each. The system uses fuzzing to quickly generate a broad input corpus and follows up with targeted branch exploration to explore paths that fuzzing struggles to reach. Findings from experiments on two C projects of different complexities show that the system did not outperform the individual techniques in terms of raw coverage, revealing limitations of the approach and opportunities for future improvement.

Essays in Financial Economics and Econometrics

Mon, 01 Sep 2025 00:00:00 GMT

Essays in Financial Economics and Econometrics Orestes, Victor M. This thesis comprises three essays in finance and econometrics. The first two essays focus on the role of credit access and liquidity in shaping real firm outcomes. The first essay examines the transmission of modern monetary policy through corporate asset markets. Exploiting quasi-experimental variation in the Central Bank of Brazil’s collateral framework and implementing a novel dynamic regression discontinuity design, it shows that monetary policy can ease expected future borrowing constraints, reduce firms’ precautionary cash holdings, and stimulate employment. The second essay analyzes how receivables financing through factoring helps firms smooth cash flows. Using a shift-share instrument and matched administrative data, it finds that cheaper liquidity leads firms to rely more on permanent labor. The third essay develops a new method for distributional inference—nonparametric quantile mixture models. This framework can be applied to financial settings such as tail risk estimation and density forecasting, as well as to causal inference when the objective is to estimate the distributional effects of interventions. It is used here to quantify the heterogeneous wage effects of a major environmental disaster. The first chapter (joint with Luis Alvarez and Thiago Christiano Silva) studies how modern monetary policy tools, which increasingly operate through corporate asset markets, affect real firm outcomes. We exploit quasi-experimental variation from the inclusion of specific corporate debt instruments in the Central Bank of Brazil’s collateral framework and implement a novel dynamic regression discontinuity design. We find that eligibility increases firms’ debt issuance, modestly lowers spreads, and reduces cash holdings, reflecting a decline in precautionary savings. These effects translate into higher employment and greater supply chain liquidity. We interpret the mechanism through the lens of segmented financial markets: by relaxing firms’ expected future borrowing constraints, the policy acts as a persistent borrowing subsidy and liquidity injection. This encourages firms to reduce cash hoarding and expand production. Using a semi-structural framework calibrated to our reduced-form estimates, we find that an induced 0.8% borrowing subsidy leads to a 1% increase in debt issuance, a 0.2% reduction in cash holdings, and a 0.4% increase in the wage bill. The second chapter (joint with Thiago Christiano Silva and Henry Zhang) shows that firms experience large increases in sales and purchases after receiving cheaper liquidity. We focus on factoring, defined as the supplier-initiated sale of receivables. In Brazil, receivables funds (FIDCs) securitize receivables for institutional investors. By assembling a novel transaction-level dataset of factoring with other credit operations for all registered firms and FIDCs, we construct a shift-share instrument for factoring financing supply based on FIDC flows. We then use a novel combination of electronic payments, trade credit, and employer-employee matched data to estimate the impacts. A flow-induced increase in receivables demand reduces firms’ factoring interest rate. In response, firms demand more permanent labor and less temporary labor. In our model, these effects arise from factoring’s purpose of reducing cash inflow volatility, helping firms match inflows to outflows, which firms otherwise achieve at an efficiency cost through substitution across labor types. The third chapter (joint with Luis Alvarez) introduces nonparametric quantile mixture models as a computationally convenient and flexible alternative to standard nonparametric density mixtures, which are widely used in Statistics and Econometrics but face significant computational and inferential challenges. We propose a sieve estimator based on a generalized method of L-moments and develop a full inferential theory. In doing so, we contribute to the statistical literature by extending a numerical bootstrap method to high-dimensional settings. As a direct application of our theory, we provide the first inference method for the distributional synthetic controls of Gunsilius (2023), a novel tool for counterfactual analysis that previously lacked formal inference procedures. We apply this method to evaluate the effects of the Brumadinho dam collapse—a large-scale environmental disaster—on the local wage distribution. The results reveal substantial heterogeneity across the distribution, with evidence of displacement effects in which median-paying jobs are replaced by lower-wage contracts. JEL Codes: C1, E4, E5, G2, G3

Machine-Learned Representations of Basis Sets and Their Application in Quantum Computational Chemistry

Mon, 01 Sep 2025 00:00:00 GMT

Machine-Learned Representations of Basis Sets and Their Application in Quantum Computational Chemistry He, Wenhao Quantum simulations of electronic structure promise to deliver significant speedups over classical methods, but remain limited by the number of qubits on near-term devices. A key strategy to reduce quantum resource requirements is to truncate the molecular Hilbert space via compact and efficient basis sets. However, most optimized basis sets either rely on predefined heuristics or require expensive classical computations, such as CASSCF orbital optimization or ℓ1-norm minimization of the Hamiltonian. In this work, we introduce a general machine learning framework for fast basis set prediction in quantum computational chemistry. Our method employs an equivariant graph neural network that outputs a Hermitian matrix encoding optimized molecular orbitals. The eigenvectors of this matrix define a transferable and efficient basis set, trained on orbitals obtained via CASSCF and Hamiltonian ℓ1 norm optimization. We evaluate our model on hydrogen chains and demonstrate that the predicted bases achieve energy accuracy and Hamiltonian sparsity comparable to orbital-optimized methods, while reducing classical preprocessing time. In addition, the predicted orbitals can be directly used as high-quality initial guesses for CASSCF calculations, further accelerating their convergence.

Signaling at the Tumor-Immune Interface in Glioblastoma

Mon, 01 Sep 2025 00:00:00 GMT

Signaling at the Tumor-Immune Interface in Glioblastoma D'Souza, Alicia D. Glioblastoma (GBM) is a devastating brain cancer, and the standard of care has not changed in over 20 years. GBM tumors are composed of a milieu of cancer cells and innate immune cells, which are co-opted by the cancer cells to promote an anti-inflammatory environment. Despite tremendous success in immunotherapy in several cancers over the past 10 years, immunotherapies have failed to show efficacy in GBM. A systems biology approach to characterizing temporal changes in tumor-immune interface of glioblastoma could illuminate new strategies to activate an anti-tumor immune response by examining changes in cell signaling and antigen presentation. In the first part of my thesis, I investigated how macrophages alter their phenotype in response to tumor co-culture and how these changes are reflected at the level of the phosphoproteome. To characterize signaling changes in distinct cell populations during co-culture, I developed a method to preserve and analyze cell-type-specific signaling using fixation. This approach enables phosphoproteomic profiling of two interacting cell types, capturing dynamic signaling events with cell-type resolution. I applied this method to study co-cultures of glioblastoma (GBM) cells and primary human macrophages. When cultured together, GBM cells induced an anti-inflammatory, immunosuppressive phenotype in macrophages, mirroring features observed in the glioblastoma tumor microenvironment. Phosphoproteomic analysis revealed that this phenotypic shift was accompanied by distinct signaling alterations in macrophages, including the upregulation of ABL kinase activity. To test this finding, I treated macrophages with an ABL kinase inhibitor and observed a reduction in the anti-inflammatory phenotype, suggesting that ABL signaling plays a role in supporting immunosuppressive macrophage polarization. Furthermore, in a mouse model of GBM, treatment with an ABL kinase inhibitor led to a reduction in the abundance of anti-inflammatory macrophages within the tumor and was associated with a modest extension of survival. In the second part, I examined changes in antigen presentation and signaling in glioblastoma tumors in response to treatment with an oncolytic virus (OV). In patient derived tumor (PDX) models in mice, mice treated with OV have increased antigen presentation, pointing to the use of OV therapy to reshape the tumor micro-environment to a more inflammatory state. Finally, tissue obtained from serial biopsies of GBM patients treated with OV shows an increase in antigen presentation and both Class I and Class II MHC protein expression. We also observed an increase in interferon alpha and interferon gamma signaling pathways as well as early induction of apoptotic pathways. These findings highlight the role of therapeutics in altering the tumor microenvironment and potentially priming it for combination immunotherapies. This thesis explores the dynamic nature of the tumor and immune compartments in glioblastoma and underscores how therapies can act on the immune compartment to promote anti-tumor activity.

SmellNet: A Large-scale Dataset for Real-world Smell Recognition

Mon, 01 Sep 2025 00:00:00 GMT

SmellNet: A Large-scale Dataset for Real-world Smell Recognition Feng, Dewei The ability of AI to sense and identify various substances based on their smell alone can have profound impacts on allergen detection (e.g., smelling gluten or peanuts in a cake), monitoring the manufacturing process, and sensing hormones that indicate emotional states, stress levels, and diseases. Despite these broad impacts, there are virtually no large-scale benchmarks, and therefore little progress, for training and evaluating AI systems’ ability to smell in the real world. In this paper, we use portable gas and chemical sensors to create SmellNet, the first large-scale database that digitizes a diverse range of smells in the natural world. SmellNet contains about 180,000 time steps of 50 substances (spanning nuts, spices, herbs, fruits, and vegetables) with 50 hours of data. Using SmellNet, we trained AI models for real-time classification of substances based on their smell alone. Our best methods leverage sequence models, contrastive learning to integrate high-resolution Gas Chromatography–Mass Spectrometry molecular data, and a new temporal difference method that identifies sharp changes in sensor readings. Our best models achieve up to 65.35% accuracy on pre-recorded data, and generalize to real-world conditions with 10.71% accuracy on nuts and 25.38% on spices in the challenging 50-way online classification task. Despite these promising results, SmellNet highlights many technical challenges in building AI for smell, including richer feature learning, on-edge smell models, and robustness to environmental changes.

Towards Zero-Shot Pretrained Models for Efficient Black-Box Optimization

Mon, 01 Sep 2025 00:00:00 GMT

Towards Zero-Shot Pretrained Models for Efficient Black-Box Optimization Meindl, Jamison Chivvis Global optimization of expensive, derivative-free black-box functions requires extreme sample efficiency. While Bayesian optimization (BO) is the current state-of-the-art, its performance hinges on surrogate and acquisition function hyperparameters that are often hand-tuned and fail to generalize across problem landscapes. We present ZeroShotOpt, the first general-purpose, pretrained model for continuous black-box optimization tasks ranging from 2 D to 20 D. Our approach leverages offline reinforcement learning on large-scale optimization trajectories collected from 12 BO variants. To scale pretraining, we generate millions of synthetic Gaussian process-based functions with diverse landscapes, enabling the model to learn transferable optimization policies. As a result, ZeroShotOpt achieves robust zero-shot generalization on a wide array of unseen synthetic and real-world benchmarks, matching or surpassing the sample efficiency of leading global optimizers, including BO, while also offering a reusable foundation for future extensions.

Temperature Characterization of Colloidal Quantum Dot Light Emitting Diodes

Mon, 01 Sep 2025 00:00:00 GMT

Temperature Characterization of Colloidal Quantum Dot Light Emitting Diodes Nguyen, Thienan D. Colloidal quantum dot light emitting diodes reveal to be promising candidates for the next generation of display technologies. Their brighter emissions, greater color purity, and higher efficiency make them highly desirable in consumer electronics. As such, research into the performance and stability of these novel LEDs are crucial for their operation in displays. These investigations are ongoing, with focused efforts on improving the operating stability through different quantum dot materials and passivation methods. However, less attention is paid in confidently understanding the fundamental relationships between current, voltage, and luminance by which these devices operate. These electrical characteristics reveal insights into the operation of these devices and the behavior of charge carriers. Additionally, temperature-dependent electrical measurements can showcase different behavior at different temperatures and deviations from the expected performance at set temperatures. Temperature dependent processes are revealed and from such, a better understanding of how the device operates is gained. In this thesis, an investigation into the temperature-dependent electrical characteristics of quantum dot light emitting diodes was conducted by measuring the current-voltage-luminance, JVL, relationships at various cryogenic temperatures. These temperatures ranged from 78K, liquid nitrogen boiling point, to 293K, room temperature. This investigation revealed the temperature dependent nature and origin of turn-on voltage, current, EQE, EQE roll-off, and hysteresis.

Uncertainty-Aware Knowledge Graph Retrieval Methods and Their Use in LLM Question-Answering

Mon, 01 Sep 2025 00:00:00 GMT

Uncertainty-Aware Knowledge Graph Retrieval Methods and Their Use in LLM Question-Answering Rich, Benjamin R. Knowledge Graph Question Answering (KGQA) encompasses a set of techniques aimed at generating accurate, interpretable responses to natural language queries posed over structured, graph-based datasets. Recent approaches to KGQA involve reducing the knowledge graph (KG) to a relevant subgraph, which is then encoded in natural language as a series of triples (subject, predicate, object) and passed to a large language model (LLM) for interpretation and answer generation. These methods have shown state-of-the-art accuracy. However, this paradigm is undermined by a critical vulnerability: the retrieval of irrelevant or erroneous facts can amplify LLM hallucinations and degrade system trustworthiness, while the reasoning process remains opaque. This thesis addresses this challenge by extending an existing stateof-the-art KGQA architecture with uncertainty-aware subgraph retrieval methods. To achieve this, we modify the retrieval component to learn the epistemic uncertainty of each candidate triple’s relevance to a given query. We implement these modifications using Bayesian methods and learn a well-calibrated approximation of the posterior distribution over triple relevance. By explicitly modeling this uncertainty, the retriever model is shown to provide a fine-grained confidence score for each piece of evidence. We expose these metrics downstream to the LLM during reasoning and evaluate whether LLMs can reason over uncertainty-related metrics to improve KGQA. We find that LLMs cannot reason effectively over uncertainties in most cases, but that agentic workflows that provide selective access to uncertainty metrics may enhance performance. We evaluate our approach against established benchmarks using HIT-rate and set-comparison accuracy metrics. Additionally, we introduce reasoning-path and statistical trust metrics derived from calibrated uncertainty scores. Our analysis reveals a significant positive correlation between path-based uncertainty metrics and the veracity of the Large Language Model’s (LLM) answers. These findings establish a robust foundation for developing uncertainty-grounded trust mechanisms in LLM-agnostic KGQA systems. As a proof of concept, a lightweight classifier trained exclusively on the LLM’s inputs and outputs demonstrates substantial predictive power in identifying correct responses. Finally, we briefly explore using uncertainty to identify out-of-distribution (OOD) queries.

Applied Compiler Optimizations for Proving Code

Mon, 01 Sep 2025 00:00:00 GMT

Applied Compiler Optimizations for Proving Code Ruiz, Ricardo The recent popularity of massively distributed, trustless systems has created a demand for cryptographic proofs: systems to prove that a piece of data is a valid output for a given program. These systems exist, but face very high runtimes for the generation of proofs. Significant effort has been invested in optimizing the prover systems, but relatively less has been focused on optimizing the code that gets read as an input. This paper proposes a new approach to optimizing prover systems by modifying the compiler to produce proof-ready code. It proposes a benchmarking framework for comparing the relative proof costs of RISC-V instructions; the resulting analyis find that shift instructions do not offer heavy savings over multiplication. The finding suggests that strength reduction, a fundamental optimization in modern compilers, can sabotage end-to-end performance. The paper proposes methods for applying this knowledge to better optimize code, leaving the door open for future researchers to continue to make code proofs more performant and accessible.

Reconstructing Cross-Species Ancestral Adeno-Associated Viruses for Enhanced Gene Therapy Delivery

Mon, 01 Sep 2025 00:00:00 GMT

Reconstructing Cross-Species Ancestral Adeno-Associated Viruses for Enhanced Gene Therapy Delivery Xie, Yuxin Adeno-associated viruses (AAV) are one of the most promising vectors for gene therapy because of their established safety, low immunogenicity, and capability to achieve sustained gene expression. However, many naturally occurring AAV variants have limitations in their potency, particularly in penetrating biological barriers like the blood-brain barrier (BBB). Additionally, their broad and nonspecific tropism can translate into suboptimal cross-species transduction efficiency and potential toxicity, complicating the clinical transition from animal model to humans. These challenges impede the use of naturally occurring AAVs for therapeutic gene delivery in many neurological disorders-such as autism spectrum disorders (ASD), Parkinson’s disease (PD), Huntington’s disease (HD)—as well as other systemic conditions like cystic fibrosis (CF). To overcome these barriers, we developed a computational framework based on ancestral sequence reconstruction (ASR) to engineer synthetic ancestral AAV capsids with the goal of enhanced targeting specificity and potency. We first validated this computational framework by replicating the previously engineered Anc80L65 capsid. Then, with 75 naturally occurring functional AAV sequences and additional experimentally screened variants exhibiting brain-targeting potency, we built an evolutionary framework. We applied multiple computational methods such as enhanced multiple sequence alignment, maximum-likelihood-based phylogenetic tree inference, and ancestral sequence reconstruction with Bayesian inference. With this methodology, we predicted several novel ancestral AAV capsid sequences at critical evolutionary nodes, particularly those representing functional transitions with potential improved blood-brain barrier penetration and CNS tropism. Our computational framework thus streamlines and accelerates the process of designing ancestral AAV variants with targeted gene therapy applications.

Intercellular flow-mediated force relaxation measurement on the three-dimensional multicellular tissue

Wed, 01 May 2024 00:00:00 GMT

Intercellular flow-mediated force relaxation measurement on the three-dimensional multicellular tissue Liu, Fan Three-dimensional (3D) multicellular tissues are prevailing over 2D monolayer or single cells; their mechanical properties like stiffness, surface tension, and viscosity have been shown to relate to diseases like fibrosis or tumor metastasis. Multicellular tissues have been traditionally modeled as a viscoelastic material due to their apparent shape rearrangement, which hardly considers the internal structure, including the extracellular matrix (ECM) and resulting intercellular water flow. These intercellular communications usually provide significant information on diseases such as tumor invasion, but immediate supporting evidence of this behavior is lacking. In this work, we investigate the bulk response of 3D multicellular tissues due to such intercellular flows and explore the related mechanism through a tailored micro-mechanics platform. Firstly, we design and establish a micro-mechanics platform based on the parallel plate compression (PPC) method. We adopt a precise micro-balance as the sensor to detect the force variation of the sample during compression. A piezo linear stage is incorporated to exert such tiny vertical displacement. Besides, a lateral microscope is designed to monitor the compression process instantaneously. This platform has proved to be applicable to various samples, including hydrogels, cell spheroids, and natural tissues or organs. Then, we propose the critical criterion, the size dependency of force relaxation time, to distinguish a material's properties, i.e., viscoelasticity and poroelasticity. For poroelastic material, the force relaxation is due to water redistribution; hence, the speed highly depends on the sample sizes. In contrast, for viscoelastic material, it is determined by the bulk material properties, thus independent of the size. We theoretically verify this criterion via Abaqus simulation and experimentally on classic poro-/visco-elastic materials with various dimensions. Next, we apply the size-dependency criterion on the 3D multicellular tissues to distinguish the poro-/visco-elasticity in this biomaterial. We take the PPC on multiple cell spheroids with different sizes through the platform. It is observed that the force relaxation times are linearly proportional to the size of all tested cell lines, demonstrating poroelasticity in our experimental time range. Intriguingly, we take tests on the natural organs of the mouse islets and find such linear correlation as well. Hence, both cultured spheroids and natural tissues are poroelastic. Finally, we explore the mechanism determining the poroelasticity inside the 3D multicellular tissues. By inhibiting the cell-cell junctions, we demonstrate the intercellular water flow through the extracellular gaps dominates this poroelastic force relaxation in the biomaterial. Further experiments show that the stiffness of the structure and the extracellular gaps inside the 3D multicellular tissues couple to contribute to the intercellular water flow, i.e., the stiffer the structure and/or the larger the gaps, the faster the water flows, thus quicker the force decays after compression. These findings highlight the fundamental role of intercellular water flow in the mechanical properties of 3D multicellular tissues. The designed micro-mechanics platform is also beneficial to research at the tissue level with micro-newton forces owing to the development of artificial organoids for early disease diagnosis and treatment.

Generating Unprecedented Extreme Scenarios with Limited Data

Mon, 01 Sep 2025 00:00:00 GMT

Generating Unprecedented Extreme Scenarios with Limited Data Chang, Kai Quantifying and predicting rare and extreme events persists as a crucial yet challenging task in understanding complex dynamical systems, ubiquitous in science and engineering. Many practical challenges arise from the infrequency and severity of these events, including the considerable variance of simple sampling methods and the substantial computational cost of high-fidelity numerical simulations. Numerous data-driven methods have recently been developed to tackle these challenges. However, a typical assumption for the success of these methods is the occurrence of multiple extreme events, either within the training dataset or during the sampling process. This leads to accurate models in regions of quiescent events but with high epistemic uncertainty in regions associated with extremes. To overcome this limitation, we introduce the framework of Extreme Event Aware (e2a or eta) or η-learning which does not assume the existence of extreme events in the available data. η-learning reduces the uncertainty even in ‘unchartered’ extreme event regions, by enforcing the extreme event statistics of a few observables during training, which can be available or assumed through qualitative arguments or other forms of analysis. This type of statistical regularization results in models that fit the observed data, but also enforces consistency with the prescribed statistics of some observables, enabling the generation of unprecedented extreme events even when the training data lack extremes therein. Theoretical results based on optimal transport offer a rigorous justification and highlight the optimality of the introduced method. Additionally, extensive numerical experiments illustrate the favorable properties of the ηlearning framework on several prototype problems and real-world precipitation downscaling problems.

A*-Decoding: Token-Efficient Inference Scaling

Mon, 01 Sep 2025 00:00:00 GMT

A*-Decoding: Token-Efficient Inference Scaling Chatziveroglou, Ioannis Inference-time scaling has emerged as a powerful alternative to parameter scaling for improving language model performance on complex reasoning tasks. While existing methods have shown strong performance gains under fixed compute budgets, there has been little focus on optimally utilizing that budget during inference. In this work, we introduce A*-decoding, a search-based inference-time strategy that builds on the A* search algorithm to optimally utilize a fixed compute budget by prioritizing high-quality reasoning paths during generation. We frame language model decoding as a structured search in a state space of partial solutions, applying the A* transition model to identify promising continuations guided by an external process supervision signal. In our experiments, A*-decoding reaches the performance levels of strong inference scaling baselines like best-of-N and particle filtering while using up to 3x fewer tokens and 30% fewer PRM passes under equivalent compute budgets. On the MATH500 and AIME 2024 benchmarks, A*-decoding enables Llama-3.2-1B-Instruct to match the performance of the 70x larger Llama-3.1-70B-Instruct, and allows Qwen3-1.7B to reach o1-like reasoning accuracy. These results highlight the power of structured search in decoding, offering an alternative to brute-force sampling or scale-driven gains. Our work demonstrates how thoughtful inference-time strategies can enhance reasoning in SLMs, pointing toward future advances in more efficient and scalable language model deployment.

U-Net Network Enhancements to Facilitate Rapid Electron Microscopy Imaging for Connectomics

Mon, 01 Sep 2025 00:00:00 GMT

U-Net Network Enhancements to Facilitate Rapid Electron Microscopy Imaging for Connectomics Varma, Vikram Imaging the structural and functional connections between cells in the brain allows neuroscientists to understand the brain by studying neuronal wiring diagrams. To automatically segment and classify images that are used in the construction of these neuronal wiring diagrams, or connectomes today, machine learning segmentation techniques require an image scanned with an electron microscope at either a slow dwell time or with small pixel sizes. However, a scalable and more rapid implementation of connectome construction has not yet been realized because of the significant cost of multi-beam electron microscopes and the relatively slow time in which connectomes can be constructed using a single-beam electron microscope. Segmented connectomes include sections that can be segmented properly with a fast scanned image as well as sections that require slow scanning for proper segmentation. Due to this fact, a potential way to enhance the time in which connectomes can be produced and segmented is to first scan samples at fast resolution and perform segmentation using a convolutional neural network, identify those areas of interest that require more detailed imaging through a learning-based error detection network, and then rescan only those identified high interest areas to produce a fused image for segmentation. The proposed thesis will analyze various machine learning methods for segmentation using the U-Net network and review proposed enhancements to the U-Net network that can better utilize electron microscopy images for construction of segmented connectomes. The successful use of fused electron microscopy images will potentially enable higher speed and lower cost electron microscopy imaging for connectomics.

Low-Temperature Germanium Waveguides for Mid-Infrared Sensing Applications

Mon, 01 Sep 2025 00:00:00 GMT

Low-Temperature Germanium Waveguides for Mid-Infrared Sensing Applications Zhang, Erin Wei Waveguide integrated devices that operate in the mid-infrared (mid-IR) wavelength range (2.5-12 µm) are used for sensing the fundamental absorption bands in a variety of molecules. Germanium (Ge) is commonly used for photodetection in the nearinfrared (near-IR) wavelength range of 1.2-1.6 µm due to its strong absorption from a 0.8 eV direct band gap. At longer wavelengths in the mid-IR range, Ge exhibits transparency that makes it a desirable waveguide material for sensing applications. Its epitaxial growth compatibility with silicon (Si) substrates makes Ge-on-Si an effective platform for mid-IR waveguides. For back-end-of-line (BEOL) integration of waveguides in sensing applications, the thermal budget limits the temperature to below 450°C. In this work, we investigated the use of h-line exposure as a commercially viable, low-cost option for patterning low temperature (LT) Ge-on-Si waveguides using direct write lithography. Waveguide dimensions for optimal confinement in single-mode transverse electric (TE) polarization at wavelengths of 3 µm and 10.4- 11.3 µm were modeled and the direct lithography process was refined. Through dose testing and adjustments to the raster direction and pixel resolution, it was found that direct write lithography lacked the resolution required for low-loss waveguides. Scanning electron microscopy (SEM) revealed inconsistent waveguide widths and sidewall roughness, and e-beam lithography was identified as the preferred lithography process. For future integration of LT-Ge in a foundry process design kit (PDK), a universal thickness of 1.7 µm was found to support single-mode waveguide operation from 3-11.3 µm wavelength.

Assessing Log-Based Coordination Systems for Managed Cloud Environments

Mon, 01 Sep 2025 00:00:00 GMT

Assessing Log-Based Coordination Systems for Managed Cloud Environments Jimenez, Gabriel The distributed systems landscape is undergoing a significant shift toward managed cloud environments, reducing the prevalence of self-hosted coordination services such as ZooKeeper. While ZooKeeper remains a proven and feature-rich solution for coordination tasks, its deployment in cloud environments can introduce component redundancy. This is because the underlying cloud platform already provides internal mechanisms to ensure coordination guarantees. This thesis investigates the design and evaluates the performance of a log-based coordination service library tailored for managed cloud environments. The proposed library removes the ensemble management overhead inherent in ZooKeeper by delegating durability and consistency responsibilities to the cloud provider’s data layer. This architectural simplification enables a modular design, allowing for tailored implementations that exploit the strengths and mitigate the limitations of a system's specified data layer. The library demonstrated feature parity with ZooKeeper for a targeted subset of coordination features, including leader election, membership tracking, and ephemeral state management. The same is noted for migration from an existing ZooKeeper-based application to this work's library, requiring minimal design changes while preserving coordination guarantees. While the results show that this design does not yet match mature coordination services in raw performance, they highlight potential avenues for further research, particularly in optimizing log-based coordination systems for the unique characteristics of cloud-managed data layers. Given the industry’s steady movement toward cloud-native infrastructure, these findings provide a foundation for future exploration into lightweight, platform-integrated coordination solutions.

Language Comprehension, Production, and Reasoning in Humans and Neural Language Models

Mon, 01 Sep 2025 00:00:00 GMT

Language Comprehension, Production, and Reasoning in Humans and Neural Language Models Eisape, Tiwalayo How closely do neural language models mirror human language processing, and what can this alignment teach us about cognition? This dissertation presents convergent evidence in comprehension, production, and reasoning that neural language models (LMs) can serve as productive instruments for understanding naturalistic human language use at scale. Studies 1-2 examine comprehension with complementary methods. First, Cloze Distillation—a novel method for aligning models with human next-word predictions—improves both language modeling and reading time prediction, demonstrating that LMs and humans make distinct, complementary predictions. Second, new methods for identifying syntactic information in LM hidden states demonstrate that models learn to implicitly represent incremental syntactic state. These probes also enable targeted interventions, allowing us to manipulate representations to resolve (or induce) temporary misinterpretations, confirming mechanistic understanding. While these studies demonstrate prediction’s role in comprehension, a complete account requires examining whether these mechanisms also shape how humans produce language in real-time. Study 3 analyzes a massive corpus of 2.3 million competitive typing events from TypeRacer.com, uncovering the first evidence of in-context predictability effects in this domain of production. Finally, Study 4 compares human and LM reasoning systematically—LMs achieve higher syllogistic reasoning accuracy than humans while still replicating several fine-grained human-like error patterns that are orthogonal to logical accuracy, including premise ordering effects. These converging findings reveal prediction as a fundamental mechanism in comprehension, production, and reasoning in both humans and LMs. While models achieve this through statistical learning rather than specialized cognitive architecture—often outperforming humans yet replicating their systematic biases—this alignment supports predictive processing theories of cognition. This work establishes LMs as scalable cognitive laboratories that can complement traditional experiments, and contributes psycholinguistically principled methods for understanding and controlling LMs.

Proof-of-Work Based Mitigation of Real Time Video DDoS Attacks

Mon, 01 Sep 2025 00:00:00 GMT

Proof-of-Work Based Mitigation of Real Time Video DDoS Attacks Echezona, Chukwuemekalum As the Internet continues to grow in size and complexity, Distributed Denial of Service (DDoS) attacks grow in size and complexity alongside it. One particularly common form of DDoS attack is the TCP SYN flood, which exploits the TCP handshake process to exhaust server resources. This thesis investigates the use of a novel proof-of-work (PoW) based mitigation method to respond to such attacks, specifically in the context of WebRTC video conferencing applications. PoW aims to shift the computational burden from the server to the client, by utilizing a hard to solve puzzle that is easily verifiable. Guided by the same evaluation framework used by the original contributors, we conducted controlled experiments using SPHERE, a national research testbed, and the open-source Jitsi Meet video conference application to simulate DDoS attacks and measure their impact on video quality metrics such as upload/download bitrate and video framerate. Our experiments involved multiple scenarios with and without active attacks and PoW mitigation activate. Results demonstrate that PoW imposes minimal overhead on legitimate clients while maintaining high efficacy when faced with the threat of a SYN Flood attack, regardless of whether the attackers do the proof-of-work before sending traffic. These findings highlight PoW as a promising low overhead mitigation method for WebRTC conference systems under the threat of DDoS attacks.

Optimizing Priority-Based Search for Lifelong Multi-Agent Path Finding

Mon, 01 Sep 2025 00:00:00 GMT

Optimizing Priority-Based Search for Lifelong Multi-Agent Path Finding Huang, Natalie The lifelong Multi-Agent Path Finding (MAPF) problem requires planning collision-free trajectories for agents operating continuously in dynamic environments. Traditional solvers such as Priority-Based Search (PBS) use fixed branching heuristics, which can be inefficient in high-congestion scenarios. This work explores how learning-based methods can improve PBS decision-making. We develop supervised learning (SL) policies trained from high-quality beam search trajectories and reinforcement learning (RL) policies learned directly through simulation, enabling adaptive branching strategies. Evaluations on warehouse-style and Kiva-style maps with varying agent densities show that learned policies can significantly boost throughput in congested warehouse layouts, while identifying scenarios where classical heuristics remain competitive. Our findings provide guidance on solver selection based on environment layout and congestion characteristics.

Learning to Interpret Language Model Diffs

Mon, 01 Sep 2025 00:00:00 GMT

Learning to Interpret Language Model Diffs Goel, Avichal Finetuning-induced changes to a model’s weights (a “model diff”) are semantically meaningful but often difficult to interpret. This makes us wonder: can we describe the content of an unknown model diff using natural language? We introduce diff interpretation training, a method that teaches a model describe its own finetuning-induced modifications. Our approach uses synthetic model diffs to train a lightweight adapter, which in turn can be applied to a compatible finetuned model to make it self-describing. Using two simple task settings, we demonstrate that our method can successfully decode model diffs into accurate natural language descriptions.

Product architectures for solar-powered drip irrigation (SPDI) systems in the Middle East and North Africa

Mon, 01 Sep 2025 00:00:00 GMT

Product architectures for solar-powered drip irrigation (SPDI) systems in the Middle East and North Africa Grant, Fiona R. To feed the growing global population, agriculture production must be intensified using existing land and resources. Sustainable agriculture intensification is particularly important in the Middle East and North Africa (MENA), the most water-stressed region in the world. Solar-powered drip irrigation (SPDI) has the potential to increase water use efficiency and reduce fossil fuel use for irrigation. Despite these benefits, SPDI adoption is limited by its high investment cost and the misalignment between farmers' risk tolerance and broader sustainability goals. Past work has explored three areas of SPDI innovation: low-pressure drip emitters, system cost optimization, and precision irrigation control. This thesis integrates previous innovations in an end-to-end design process to generate SPDI architectures that are accessible to resource-constrained farmers. A market study was conducted to understand farmers' priorities and constraints and articulate SPDI value propositions for the target users. Stakeholder surveys were conducted in Jordan and Morocco for farms ranging from 1–130 hectares. Three market segments were identified, grouping farmers who face similar economic and knowledge barriers. While farmers generally prioritized irrigation reliability and low system costs, the observed variety in farm size, production volume, and technical expertise suggested that SPDI architectures must be tailored to each market segment. This thesis proposes an energetic framework that captures system parametric relationships to identify feasible SPDI design trade-offs. The optimized solar power systems were 14%–80% less expensive than conventionally-sized designs. Despite significant changes to the hydraulic operating parameters, the proposed SPDI architectures were as reliable as existing systems. For farms with long irrigation times, it was optimal to pair low-pressure drip emitters with an irrigation schedule that tracks the daily solar profile, termed “solar profile matching” (SPM), to maximize direct solar power use. The SPM schedule reduced system cost by minimizing the battery capacity. An economic analysis demonstrated that the optimal SPDI designs could be made cost-competitive with grid power through SPDI retrofit subsidies, which some local governments already support. Researchers and industry professionals could use the energetic framework and techno-economic analysis presented in this thesis to inform system design and policy decisions and promote SPDI adoption. Finally, this work created guidelines for designing a precision irrigation controller in resource-constrained markets. A controller was conceptualized to implement the SPDI-SPM architecture. The controller functional requirements and design specifications were iteratively defined with stakeholders, and a prototype was tested on two farms in the MENA region. The controller reduced water and energy use by up to 44% and 43%, respectively, while maintaining crop yield. However, the controller relied on battery power to execute the irrigation schedule. A yield loss sensitivity analysis found that using 72%–79% of the available solar energy on average, an increase of about 40% from the experiment SPM schedules, would have been sufficient to reliably irrigate with solar alone. The results suggest that, with software modifications, the proposed controller could eliminate the need for a battery and enable low-cost SPDI systems. If adopted, the proposed controller could make sustainable irrigation practices more accessible to farmers.

Approximate L² Error Control by Solution Post-Processing for Finite Element Solutions of PDEs with Higher-Order Adaptive Methods

Mon, 01 Sep 2025 00:00:00 GMT

Approximate L² Error Control by Solution Post-Processing for Finite Element Solutions of PDEs with Higher-Order Adaptive Methods Botto Tornielli, Marcos Julian With the substantial computing resources available today, computational fluid dynamics simulations allow scientists and engineers to simulate physical problems very accurately. However, achieving this accuracy requires a sufficiently refined computational mesh, which is a primary driver for the high cost of complex simulations. Mesh adaptation methods provide an automated way to determine the regions where a mesh needs the most refinement and generate a new mesh that efficiently targets these regions. In this thesis, we build on previous work in a posteriori error estimation and mesh adaptation for finite element methods to propose a new mesh adaptation method based on L² error control by solution post-processing. A key feature of our method is its natural extension to higher-order discretizations while providing a problem-independent adaptation methodology. Problem-independent adaptation methods do not depend on specific information about the partial differential equation (PDE) problem being solved, and can therefore be applied to a wide range of problems without modification. We present numerical results applying the approximate L² error control method to a two-dimensional advection-diffusion problem with anisotropic features. These results demonstrate the proposed method’s ability to generate well-adapted anisotropic meshes for solutions with polynomial orders 1, 2, and 3. We also apply the approximate L² error control method to a more complex two-dimensional Reynolds-Averaged Navier-Stokes problem with turbulent flow over a flat plate. We compare the convergence of the drag coefficient and the characteristics of adapted meshes obtained with the proposed method and with an output-based adaptation approach. As expected, the approximate L² error control method is not as effective as the output-based approach in reaching a converged drag coefficient value, but it nevertheless demonstrates the ability to effectively control the approximate L² error in the Mach field.

Advancing Ubiquitous Tactile Sensing through Comprehensive Tooling for Resistive Matrix-Based Sensors

Mon, 01 Sep 2025 00:00:00 GMT

Advancing Ubiquitous Tactile Sensing through Comprehensive Tooling for Resistive Matrix-Based Sensors Murphy, Devin Resistive matrix-based tactile sensors offer a scalable and intuitive approach to capturing human-environment interactions, yet deploying them in real-world systems remains challenging because they must remain portable, adaptive, and long-lasting. This thesis presents the WiReSens Toolkit, an open-source hardware and software platform for developing resistive tactile sensing systems that meet the demands of real world applications. The toolkit features adaptive hardware for interfacing with resistive sensors and a web-based GUI that mediates access to otherwise complex functionality, including 1) multi-device programming and wireless visualization across three distinct communication protocols 2) autocalibration methods for adaptive sensitivity and 3) intermittent data transmission for low-power operation. As a use case for the toolkit, the thesis then introduces a method for the automatic design and fabrication of custom tactile sensing gloves using flexible printed circuit boards (FPCBs), enabling rapid, scalable production. Together, these contributions lower barriers to adoption and support broader exploration of tactile sensing in HCI, robotics, and ubiquitous computing.

Topics in Geometric Machine Learning

Mon, 01 Sep 2025 00:00:00 GMT

Topics in Geometric Machine Learning Tahmasebi, Behrooz Recent advances and the widespread adoption of neural networks have revolutionized machine learning and artificial intelligence. These developments demand learning paradigms capable of processing data from diverse applications and sources. In structured domains such as molecules, graphs, sets, and 3D objects, as well as fields such as drug discovery, materials science, and astronomy, models must account for data structures. The emerging field of geometric machine learning has gained attention for enabling neural networks to handle geometric structures, unlocking novel solutions across scientific disciplines. Despite recent advances, theoretical gaps remain. This thesis aims to address these gaps by studying the benefits and limitations of leveraging geometric structures and symmetries in data. We explore sample complexity, generalization bounds, hypothesis testing for the presence of symmetries in data, time complexity of learning under symmetries, and regularization and optimization in symmetric settings. The goal is to build a robust theoretical framework that validates recent successes and sheds light on unexplored aspects, fostering future progress in geometric machine learning.

Quantization Methods for Matrix Multiplication and Efficient Transformers

Mon, 01 Sep 2025 00:00:00 GMT

Quantization Methods for Matrix Multiplication and Efficient Transformers Savkin, Semyon We study quantization in Machine Learning. First, we introduce NestQuant — a technique for quantization of matrix products and post-training quantization of LLMs. Beyond reducing the memory footprint, quantization accelerates inference, as the primary bottleneck during autoregressive generation is often the memory bandwidth. NestQuant leverages two nested lattices to construct an efficient vector codebook for quantization, along with practical encoding and decoding algorithms. The approach is grounded in recent theoretical work that characterizes the optimal rate–distortion trade-off for matrix products. Empirically, on Llama-3-8B, it reduces the perplexity gap between full-precision and quantized models by more than 55% relative to the current state-of-the-art technique (SpinQuant). Second, we investigate data-domain quantization for RF signals. We propose a tokenized transformer for source separation that discretizes RF waveforms into learned tokens and operates directly on the resulting sequences, outperforming strong convolutional baselines. Together, these contributions connect information-theoretic limits with deployable systems: structured vector quantizers accelerate LLM inference and enable competitive discrete representations for RF tasks.

Improving Data-Driven Contact Localization and Force Estimation for Barometric Tactile Sensors

Mon, 01 Sep 2025 00:00:00 GMT

Improving Data-Driven Contact Localization and Force Estimation for Barometric Tactile Sensors Chun, Ethan Barometric tactile sensors offer a cheap, robust, and customizable means for robots to perceive the world. Central to their operation are models that extract useful information from the sensors’ raw pressure readings. In this work, I focus on improving data-driven methods for single-point contact localization and force estimation using a previously presented three-quarter sphere barometric tactile sensor. To allow modeling of time-dependent effects in the sensor material, I introduce a multi-threaded data collection system that captures ground truth contact and sensor data at exactly 100 Hz. I construct both feed-forward and recurrent networks using this data, finding that a recurrent network achieves a 15% lower mean absolute error for angular contact localization on the sphere compared to prior methods. The recurrent architecture’s computational efficiency ensures that the architecture can still run within the constraints of the sensors’ microcontroller. Despite this improvement, I find that more expressive models such as LSTMs tend to overfit on the collected data and physical phenomena observed during deployment were not well represented by the training metrics. To better understand the extent that these data-driven methods alone can improve sensor performance, I shift focus away from the modeling and analyze the physical sensor instead. I find that viscous effects in the sensor can render the prediction task unlearnable without historical data and that thermal effects introduce a train-test distribution shift. Finally, I discuss design criteria for a theoretical future barometric tactile sensor that may mitigate the effects found during my modeling and analysis.

Probabilistic Programming over Heterogeneous Language and Hardware Targets

Mon, 01 Sep 2025 00:00:00 GMT

Probabilistic Programming over Heterogeneous Language and Hardware Targets Rojas Collins, Elias G. Modern probabilistic programming applications, from large-scale Bayesian inference to real-time decision making, require both the expressiveness of CPU-oriented languages such as Gen.jl and the massive parallelism of GPU-backed array languages such as GenJAX, yet existing platforms force users to trade modeling flexibility for performance. This thesis introduces GenUflect, a metalanguage that embeds multiple Gen-compatible dialects inside a single program, allowing each sub-component to run on the most appropriate language and hardware target while preserving Gen’s programmable-inference interface. GenUflect extends Gen’s dynamic-modeling language with the @union, @vmap, @amortize, @amortize≤, and @runtime_union combinators; these macros compile at build-time (or justin-time) to autonomous generative functions written in the target dialect, link them through a lightweight FFI layer, and manage cross-device data via zero-copy MirrorArrays and lazily materialized traces. The resulting programs remain sound by construction because each foreign subtrace is itself a valid Gen generative function. Empirical studies demonstrate that this hybrid approach yields large practical gains. On a split linear-vs-sinusoidal regression task, GenUflect matches pure GenJAX throughput while running higher-order control logic on the CPU, and is up to two orders of magnitude faster than a pure Gen implementation for datasets of 105 points. In a collapsed-Gibbs sampler for a Dirichlet-process mixture model, GenUflect’s elastic allocation (@amortize≤) lets vectorized GPU kernels adapt to a growing number of clusters; the same inference that takes over an hour in Gen executes in seconds with GenUflect. A probabilistic inverse-graphics pipeline further showcases how heterogeneous submodels can cooperate seamlessly within unified inference code. By coupling language interoperability with automated data movement and compile-time code generation, GenUflect bridges the gap between flexibility and speed, enabling scalable, expressive probabilistic programs that natively exploit both CPUs and accelerators.

Under-Coverage of Double Machine Learning Due to Implementation Choices

Mon, 01 Sep 2025 00:00:00 GMT

Under-Coverage of Double Machine Learning Due to Implementation Choices Siegmann, Charlotte B. Double ML estimators can estimate coefficients of interest with far fewer functional form assumptions than linear econometric methods. However, DML requires researchers to make a range of implementation choices, including the selection of the function class, the random seed, and hyperparameter configurations. While asymptotic theory suggests these choices should not affect final estimates, we show that for 10 economic analyses (8 of them published and peer-reviewed), implementation choices affect the results. In half of the datasets, different implementation choices even change the interpretation of findings between negative, null, or positive effects. We link these results to a framework for empirically assessing the performance of machine-learning-based estimators, focusing on precision, coverage, and susceptibility to manipulation. This is meant to complement asymptotic theory. We demonstrate that the coverage of DML confidence intervals is too low—placing an upper bound of 48% on the expected coverage of conventional 95% confidence intervals for published DML economics papers. We show that in the status quo, the susceptibility of DML to manipulation by researchers is high, but propose ways to mitigate this susceptibility.

Optimizing Irreversible Perturbations of the Unadjusted Langevin Algorithm

Mon, 01 Sep 2025 00:00:00 GMT

Optimizing Irreversible Perturbations of the Unadjusted Langevin Algorithm Zhu, Qianyu Julie A central task in Bayesian inference and scientific computing is to compute expectations with respect to probability distributions that are only known up to a normalizing constant. Markov chain Monte Carlo (MCMC) methods, and in particular Langevin dynamics, provide a powerful framework for this task by constructing stochastic processes that converge to the target distribution. However, practical implementations face two challenges: slow mixing when the target distribution is anisotropic or multimodal, and persistent discretization bias introduced by numerical schemes. This thesis investigates irreversible perturbations of overdamped Langevin dynamics, aiming to accelerate mixing while controlling discretization error. Irreversible perturbations introduce skew-symmetric drift terms that preserve the target distribution while inducing rotational flow, thereby enhancing exploration. Although prior work has established their benefits in continuous-time settings, the impact of discretization and the design of optimal perturbations for discrete-time algorithms remain open problems. We develop a framework for optimizing constant (position-independent) irreversible perturbations in the Unadjusted Langevin Algorithm (ULA). Our approach balances two competing objectives: maximizing the spectral gap of the continuous dynamics to accelerate convergence, and minimizing discretization error that drives estimation bias. Motivated by this, we introduce new criteria that jointly evaluate bias and efficiency, and we show how these criteria identify perturbations that improve performance beyond existing constructions. Theoretical analysis is complemented by numerical experiments on Gaussian and nonGaussian targets. These experiments demonstrate that appropriately designed irreversible perturbations can reduce mean-squared error without sacrificing stability, while poorly chosen perturbations can degrade performance. The results highlight the importance of geometry-aware design and motivate systematic optimization strategies for irreversible perturbations. Overall, this work extends the theoretical and practical understanding of irreversible Langevin dynamics, bridging the gap between continuous-time spectral analysis and discrete-time numerical performance. It provides principled tools for constructing efficient MCMC samplers, with potential applications in high-dimensional Bayesian inference and modern machine learning.

Single Camera Motion Compensated Viewpoint Shift

Mon, 01 Sep 2025 00:00:00 GMT

Single Camera Motion Compensated Viewpoint Shift Snowdon, Adam Eye contact is a necessary tool for human connection and in most video conferencing situations, eye contact is not possible. Standard laptop and webcam configurations position the camera at the top of the screen, meaning that when the user looks at other people’s faces in the center of the screen, the camera captures the user looking downward, creating the impression of poor eye contact for remote participants. Solutions involving 3D modeling of the face to synthesize a gaze-corrected view have been explored and exist but are too computationally costly for most personal computers. To address this computational challenge, we draw inspiration from 2D frame interpolation techniques to synthesize a virtual camera view that repositions the user’s apparent gaze toward the camera. Our method uses a single camera located at the top of the user’s screen and requires only a brief setup period. Assuming there is only one user, our approach creates a virtual camera view that transforms the user’s viewpoint from the screen center to the camera position, enabling more realistic eye contact in video conference calls.

A Framework for 3D Mouse Brain Reconstruction: RNA-based Stitching of Adjacent Tissue Slices and Co-Registration of Multimodal Imaging Data

Mon, 01 Sep 2025 00:00:00 GMT

A Framework for 3D Mouse Brain Reconstruction: RNA-based Stitching of Adjacent Tissue Slices and Co-Registration of Multimodal Imaging Data Pan, Jessica N. Mapping the brain’s complex neural networks requires tracing the long-distance pathways of individual axons, a task that demands a comprehensive 3D reconstruction of the brain. Recently, spatially resolved transcriptomics (SRT) methods enable the study of gene expression and biomolecule distribution in each neuron in its spatial context, opening the door to more thoroughly investigating cell-cell interactions between neurons. However, SRT methods are limited to slices of tissue; therefore, computational alignment is essential to reconstruct a cohesive 3D volume while correcting for both batch effects and inherent sample variability. This thesis presents a novel framework that addresses these challenges through three primary contributions. First, a memory-efficient, non-referenced-based algorithm was developed to align the superficial surfaces of adjacent, high-resolution tissue slices. Second, these surface transformations were interpolated through the tissue slices on a proof-of-concept dataset of three adjacent slices. Third, methods for co-transforming fluorescent protein imaging data were explored to fully resolve the cell boundaries between neurons. These three methods are necessary steps towards creating a fully-resolved, multimodal 3D model of the brain.

Optimizing Non-Convex Objectives to Plan More Optimal Motion for Manipulators

Mon, 01 Sep 2025 00:00:00 GMT

Optimizing Non-Convex Objectives to Plan More Optimal Motion for Manipulators Garg, Shruti Non-convex optimization is essential to tackle increasingly complex and practical problems in kinematic motion planning. Although introducing non-convexity often sacrifices guarantees of feasibility and optimality–making solutions more susceptible to local minima or failure to converge–many robotic systems and tasks are non-convex by nature, necessitating at least somewhat non-convex formulations. In this thesis, we aim to mostly constrain non-convexity to the objective. This optimization structure helps preserve certain feasibility guarantees in theory and usability in practice while enhancing optimality of solutions, even if global optimality is not achieved. In the first chapter, we demonstrate the effectiveness of non-convex objectives in scenarios where motion planning involves a non-convex parameterization of the configuration space. We keep constraints strictly convex, with the non-convexity quarantined to the objective. This structure guarantees a feasible solution given a feasible initial guess. We primarily use our method to post-process Graphs of Convex Sets solutions in three domains: constrained bimanual motion, motion with guaranteed non-collision, and planning in SO(3). In each case, the non-convex objective compensates for distortion introduced by the parameterization, resulting in more efficient and natural motion. In the second chapter, we propose teleoperation scheme with full-body motion planning for non-holonomic mobile manipulators. Our key contribution is a Differential Inverse Kinematics (DiffIK) formulation that crafts non-convex objectives to avoid singularities and joint limits leading to more robust feasible motion. Unlike before, the constraints are not strictly convex, so the optimization has no guarantees of feasibility. However, we mitigate the non-convexity in the constraints as much as we can by linearizing around the robot’s current position and approximating the highly non-convex non-holonomic constraint. We explore multiple formulations for singularity avoidance and empirically demonstrate that integrating these objectives into DiffIK improves motion quality for teleoperation for the RBY-1 robot.

CableSplat: Optimized 3D Gaussian Splatting for 1D Deformable Pose Estimation

Mon, 01 Sep 2025 00:00:00 GMT

CableSplat: Optimized 3D Gaussian Splatting for 1D Deformable Pose Estimation Pai, Sameer A key challenge in the robotic manipulation of deformable objects is the lack of accurate and efficient systems for estimating their pose in real-time, especially in the presence of occlusion. In this thesis we propose CableSplat, a novel non-parametric method leveraging 3D Gaussian Splatting to estimate the pose of a linear deformable object given RGB images of the object from multiple viewpoints. To facilitate the evaluation of the performance of this method, we develop both simulated and real-world pipelines to collect calibrated and segmented recordings of cables undergoing various manipulations and transformations. We find that our method is consistently able to estimate cable pose to within an average error of ∼2.5mm across simulated tasks. Furthermore, performance on a scene reconstruction metric drops only slightly between simulated and real-world data, suggesting high-fidelity state estimation even in the real world. CableSplat is therefore a promising candidate for the extension of existing manipulation systems to deformables.

scPhen: Single-Cell Phenotype Predictor for Alzheimer’s Disease

Mon, 01 Sep 2025 00:00:00 GMT

scPhen: Single-Cell Phenotype Predictor for Alzheimer’s Disease Guo, Sophie J. Advances in artificial intelligence (AI) and generative AI for representation learning have transformed our ability to model complex biological systems. Single-cell RNA sequencing (scRNA-seq) provides unprecedented resolution into cellular heterogeneity, offering a powerful substrate for modeling disease circuitry. However, predicting patient-level phenotypes from scRNA-seq remains challenging due to limited sample sizes, variable cell counts, and the computational burden of modeling long-context dependencies. We present scPhen, a flexible, parametric deep-learning framework for phenotype prediction from single-cell transcriptomic data, applied here to Alzheimer’s disease (AD) as a paradigm of complex, heterogeneous pathology. scPhen consists of a cell embedding module and a patient embedding module, designed to capture both fine-grained molecular patterns and higher-order cell–cell relationships. The framework supports multiple architectural backbones, including Transformers, Graph Neural Networks (GNNs), and state-space models such as Mamba, Mamba2, and BiMamba2, allowing exploration of tunable components for optimized performance. Across classification and regression tasks, state-space models, and in particular BiMamba2, demonstrated superior predictive accuracy and computational efficiency compared to Transformer-based and hybrid approaches. We further integrated attention-based multiple instance learning to enable variable cell counts per patient and to prioritize phenotype-informative cellular subsets. Interpretability analyses using Integrated Gradients and cell-level attention scores revealed gene programs and cell populations associated with AD progression, highlighting known neuroinflammatory signatures and suggesting novel molecular targets. By unifying cutting-edge sequence modeling architectures with scalable single-cell analysis, scPhen provides a generalizable, high-resolution approach to phenotype prediction. While demonstrated here in AD, this framework is readily extensible to other complex diseases and multi-modal cellular datasets, bridging computational innovation and biological discovery.

Predicting Task Functional Localizers Using Naturalistic fMRI

Mon, 01 Sep 2025 00:00:00 GMT

Predicting Task Functional Localizers Using Naturalistic fMRI Wilke, Jordan Functional magnetic resonance imaging (fMRI) data collected during naturalistic stimuli has shown promise for predicting individual traits, biomarkers of disease and functional brain localizations, potentially offering advantages over traditional resting-state approaches. This study investigated the use of interpretable deep learning models to predict demographics and functional task localizer activations from fMRI time-series data collected while participants viewed naturalistic stimuli. Using the data of 143 subjects from the Human Connectome Project, I analyzed 7T fMRI scans from participants watching movies to predict sex, age, and functional localizer activations across multiple cognitive tasks. I employed state-of-the-art machine learning architectures, including DICE and Glacier models, specifically chosen for their interpretable design features that build directed connectivity matrices and produce weighted temporal attention maps. These models aimed to capture dynamic brain activity patterns while maintaining the ability to understand which temporal features drive predictions. The results successfully reproduced previous findings for sex classification but showed poor performance for age prediction, with correlations ranging from -0.175 to 0.243. For functional localizer predictions, models initially appeared to achieve high performance with some specific contrasts having correlations around 0.9 and Dice scores generally above 0.6. However, detailed analysis revealed that these models were primarily predicting group averages rather than learning meaningful inter-subject variability, as evidenced by chance-level subject identification accuracy. This finding contrasts with previous works that demonstrated successful prediction of individual differences in functional localizations. The failure to capture inter-subject variability represents a significant limitation, as individual differences in functional regions of interest are crucial for applications such as pre-surgical mapping and disease prediction. My findings suggest that predicting from raw fMRI time-series may require different approaches than those used here, with preprocessed functional connectivity matrices showing promising results, and highlight the importance of sufficient training data to separate signal from noise when learning directly from naturalistic stimuli. Despite these challenges, this work establishes important methodological foundations and identifies key limitations that must be addressed in future research combining naturalistic stimuli with machine learning for fMRI prediction tasks. The findings emphasize the need for models that can capture individual functional differences while maintaining the interpretability necessary for understanding how naturalistic stimuli drive brain-based predictions.

Probabilistic Programming with Low-Level, High-Performance GPU Programmable Inference

Mon, 01 Sep 2025 00:00:00 GMT

Probabilistic Programming with Low-Level, High-Performance GPU Programmable Inference Chung, Karen GPU-compatible probabilistic programming languages (PPLs) have enabled high-performance, data-parallel programmable inference. However, these systems face fundamental trade-offs between expressiveness and performance, as their GPU code generation is automated and black-boxed, limiting optimization opportunities and imposing restrictions on program expressivity. This thesis introduces GenCUDA, a probabilistic programming system that addresses this limitation by embedding the CUDA GPU programming language directly into a C++/CUDA frontend, enabling GPU programmable inference with fine-grained control over runtime and memory profiles. GenCUDA extends the Gen probabilistic programming architecture by providing a dynamic modeling language (DML) that allows users to write performance-critical sections of generative functions as CUDA kernels while maintaining automatic trace management and the generative function interface (GFI). The system supports both sequential and parallel execution contexts through specialized effect handlers that seamlessly compose CPU and GPU code paths. Key technical contributions include: (1) a high-performance GPU distributions library achieving 10-100× speedups over TensorFlow-Probability, (2) memory-efficient trace management via template-optimized parallel effect handlers, and (3) vectorized generative functions that enable massive parallelization of inference algorithms. We demonstrate GenCUDA’s capabilities through comprehensive benchmarks on inference algorithms applied to diverse models including factor graphs, mixture models, and Hidden Markov Models. Results show significant performance improvements over JAX-based implementations: up to 3× speedup for importance sampling on a hierarchical model, 5.7× speedup for parallel Gibbs sampling on factor graphs, and memory efficiency improvements for large-scale mixture models supporting up to 6× as many clusters compared to existing frameworks’ limits. The system maintains the composability and expressiveness of probabilistic programming while unlocking GPU performance optimization techniques such as kernel fusion and memory hierarchy exploitation that are inaccessible to higher-level frameworks. GenCUDA demonstrates that embedding low-level GPU programming within automated probabilistic inference workflows can achieve both performance gains and algorithmic expressivity without sacrificing the modularity of probabilistic programming paradigms.

Simplifying Equivariant GPU Kernels through Tile-based Programming

Mon, 01 Sep 2025 00:00:00 GMT

Simplifying Equivariant GPU Kernels through Tile-based Programming Kotak, Mit E(3)-equivariant neural networks have demonstrated success across a wide range of 3D modeling tasks. Until recently, they were bottlenecked due to their high memory and wall-time requirements. In this thesis we first provide an overview of recent GPU kernel efforts by both academia and industry that address this issue. These approaches tradeoff performance for engineering complexity, while still being algorithmically bottlenecked at 10 % GPU utilization. We instead trade off engineering complexity for performance. This not only lowers the barrier to GPU programming but also builds an abstraction layer to reason about future algorithmic innovations that can improve GPU utilization. Our kernel 𝐵3, based on the tiling- optimizations in just 100 lines of PyTorch-like code. We explore the performance-simplicity tradeoff with two case studies and demonstrate the practicality of our kernel workflow through downstream integration with a production model. We hope this work serves as inspiration to broaden and deepen existing equivariant kernel efforts.

Chemical exposures in drinking water: contaminant analysis and physicochemical behavior

Mon, 01 Sep 2025 00:00:00 GMT

Chemical exposures in drinking water: contaminant analysis and physicochemical behavior Bugher, Nicolette A. Environmental chemical exposures pose an understudied risk to human health. The quality and accessibility of data on occurrence in the environment and physicochemical behavior of industrial chemicals are integral for accurate exposure risk assessment. In this dissertation, analytical chemistry techniques were developed and leveraged to characterize human exposures to contaminants in drinking water and improve methods for assessing such risks. The occurrence of organic industrial pollutants in domestic well waters was investigated, with a particular focus on the impacts of region-specific industrial activity (e.g., hydraulic fracturing), legacy pollution sites (e.g., Superfund sites), and geochemistry. The exposure risk to water contaminants of domestic well users was further interrogated by evaluating trends in contaminant concentrations resulting from the implementation and maintenance of in-home water treatment devices. The results show widespread, low-dose mixtures of organic pollutants, where the efficacy of removal by in-home water treatment varied by water contaminant class and maintenance frequency. Additionally, analytical methods were optimized to quantify a group of organic water contaminants (i.e., probable carcinogens, N-nitrosamines), improving method sensitivity and critically identifying false-positive interferences. Finally, methods were evaluated and deployed for the determination of physicochemical properties of N-nitrosamines. The results of which demonstrate gaps in existing experimental data, provide a valuable methodological intercomparison (two experimental and two computational approaches), and contribute novel partitioning data. This dissertation addresses gaps in occurrence data, analytical method sensitivity, and reliability of physicochemical parameters for risk assessment. The combination of method development and implementation enables the study of exposures to water contaminant mixtures at health-relevant concentrations, representative of prevalent exposure pathways.

From coarse fate choice to precise pattern: post-mitotic progenitor targeting

Mon, 01 Sep 2025 00:00:00 GMT

From coarse fate choice to precise pattern: post-mitotic progenitor targeting Nie, Mel F. Planarians possess remarkable regenerative abilities, driven by pluripotent stem cells called neoblasts. While neoblasts are known to give rise to progenitor cells that form various tissues, whether and the extent to which these progenitors migrate across the animal remains unclear. Irradiation experiments eliminate all neoblasts outside shielded areas, allowing for the visualization of cell migration from the remaining neoblasts, but irradiated animals may not reflect homeostatic progenitor migration patterns. To address this, 5-ethynyl-2’-deoxyuridine (EdU) labeling and plug transplant techniques were used to trace progenitor movement in non-irradiated planarians. Using whole-mount fluorescence in situ hybridization (FISH) and the quantification of EdU-labeled cells, this study demonstrates that progenitor cells are capable of migrating long distances and exhibit a pronounced anterior bias in their movement and integration.

Optimizing Data Layouts for Evolving Cloud Table Storage

Mon, 01 Sep 2025 00:00:00 GMT

Optimizing Data Layouts for Evolving Cloud Table Storage Sudhir, Sivaprasad Modern data analytics platforms increasingly adopt disaggregated architectures, storing data in cost-effective cloud object stores. While this approach enables a clean separation of concerns, allowing each layer to be independently managed and scaled, it introduces significant performance bottlenecks due to expensive data movement. Effective data layouts, which organize data to minimize unnecessary data reads, are thus critical to achieving high query performance. However, existing techniques typically rely on manually specified layouts, collect limited metadata, or lack mechanisms to dynamically adapt to changing data and workloads. This thesis investigates adaptive, metadata-rich, expressive data layouts for cloud table storage. First, we introduce Pando, a correlation-aware layout technique that leverages rich metadata on query predicates to significantly improve data skipping. Next, we propose CopyRight, a partial replication strategy that selectively replicates subsets of data and optimizes each replica differently, efficiently serving heterogeneous query patterns. Finally, we describe Self-Organizing Data Containers (SDCs), a practical table storage layer for the cloud that incrementally reorganizes complex data layouts based on changes in data and workload distributions.

Development of Ensemble Strategies for Generalization in Deepfake Image Detection

Mon, 01 Sep 2025 00:00:00 GMT

Development of Ensemble Strategies for Generalization in Deepfake Image Detection Wagh, Rohan M. The growing accessibility of generative models has enabled the rapid proliferation of deepfake content, posing significant challenges in image-based biometric security and media authenticity. In this thesis, six diverse facial deepfake image datasets are assembled, and four modern detection models are evaluated in a cross-domain scenario. We observe that individual models fail to generalize to images generated by techniques outside the scope of their training data. This often hinders the applicability of a single model in real-world deepfake detection. This thesis proposes ensemble strategies as a means of addressing this lack of generalization. We find that the ensemble models outperform individual models in classifying deepfake images, particularly in terms of accuracy and recall. An exhaustive evaluation of combinations of models shows that ensembles of similar models provide limited benefit, whereas ensembles of complementary models lead to significant improvements in classification performance. Ensembling models based specifically on accuracy and recall metrics also produces models that lower the rate of more harmful false negative predictions. This work highlights the value of ensemble models in improving generalization across diverse image families and provides a framework for building robustness in real-world deepfake detection systems.

The effect of settlements on the stresses in building frames

Tue, 01 Jan 1935 00:00:00 GMT

The effect of settlements on the stresses in building frames Granberg, Robert J. Thesis: B.S., Massachusetts Institute of Technology, Department of Building Engineering and Construction, 1935; Includes bibliographical references.

Irradiation grafting of styrene onto dacron fibers and films

Thu, 01 Jan 1959 00:00:00 GMT

Irradiation grafting of styrene onto dacron fibers and films Schnetzer, L. J.; Hendren, J. W. Thesis: B.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1959; Includes bibliographical references (leaves 43-44).

An investigation of sound transmission irregularity in a one dimensional enclosure

Sat, 01 Jan 1949 00:00:00 GMT

An investigation of sound transmission irregularity in a one dimensional enclosure Foster, Isaac C. Thesis: B.S., Massachusetts Institute of Technology, Department of Physics, 1949

Deposition and characterization of very low pressure CVD silicon/silicon-germanium heteroepitaxial structures

Wed, 01 Jan 1992 00:00:00 GMT

Deposition and characterization of very low pressure CVD silicon/silicon-germanium heteroepitaxial structures Tsai, Curtis. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 1992; Includes bibliographical references (leaves 135-146).

An experimental study of the law of parity conservation in electromagnetic interactions.

Fri, 01 Jan 1965 00:00:00 GMT

An experimental study of the law of parity conservation in electromagnetic interactions. Hegblom, Edwin Richard. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Physics, 1965

Analysis of a dynamic sales call policy model.

Mon, 01 Jan 1968 00:00:00 GMT

Analysis of a dynamic sales call policy model. Karash, Richard Ivan. Thesis: B.S., Massachusetts Institute of Technology, Department of Physics, 1968; Bibliography: leaf 97.

MEKIN numerical modeling and simulation of the SPERT-III E-core transient tests

Tue, 01 Jan 1980 00:00:00 GMT

MEKIN numerical modeling and simulation of the SPERT-III E-core transient tests Tan, Lip-Bu. Thesis: M.S., Massachusetts Institute of Technology, Department of Nuclear Engineering, 1980; Includes bibliographical references.

Comparative tests of the Boston Elevated Co's surface cars

Mon, 01 Jan 1912 00:00:00 GMT

Comparative tests of the Boston Elevated Co's surface cars Jones, Philip C.; Katsainos, Nicholas M. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1912

Rules for ring closure and aspects of organolithium chemistry

Tue, 01 Jan 1980 00:00:00 GMT

Rules for ring closure and aspects of organolithium chemistry Dupont, William Alan. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Chemistry, 1980; Vita.; Includes bibliographical references.

Dracut nickel ore ; Geology and concentration, ore no. 2592

Sun, 01 Jan 1905 00:00:00 GMT

Dracut nickel ore ; Geology and concentration, ore no. 2592 Burton, Eugene.; Spalding, William Livingston. Thesis: B.S., Massachusetts Institute of Technology, Department of Mining Engineering and Metallurgy, 1905

Evaluation of recirculating well technology with a cost comparison to pump and treat technology for containment of the CS-10 contaminant plume at the Massachusetts Military Reservation

Wed, 01 Jan 1997 00:00:00 GMT

Evaluation of recirculating well technology with a cost comparison to pump and treat technology for containment of the CS-10 contaminant plume at the Massachusetts Military Reservation Smith, Mathew D. (Mathew Darin) Thesis: M. Eng., Massachusetts Institute of Technology, Department of Civil and Environmental Engineering, 1997; Includes bibliographical references (leaves 43-45).

The crystallization of sucrose

Tue, 01 Jan 1929 00:00:00 GMT

The crystallization of sucrose Brown, Ernest K. Thesis: M.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1929; Includes bibliographical references (leaf 81).

Systems of Visualization for Musical Futures

Mon, 01 Sep 2025 00:00:00 GMT

Systems of Visualization for Musical Futures Naseck, Perry This thesis investigates how large-scale visual systems can communicate the presence, agency, and foresight of improvising musical agents–human and AI–during live performance. We propose a framework for manifesting AI collaborators on stage through five principles: musical transparency, live improvisational reactivity, demonstrated virtuosity, communication for collaboration, and visual fit. Two public performances operationalize these ideas: an addressable-light sculpture that renders harmonic space, and a stage-sized kinetic sculpture built from novel, low-cost Generic Pan Tilt fixtures that visualize the AI’s planned “musical futures.” The latter combines a real-time, MIDI-conditioned, Transformer-based hand-motion model with deterministic, pattern-based mappings that signal states such as resting and regeneration. Audience surveys indicate that viewers perceived links between musical turns and kinetic gestures while requesting clearer explanatory cues. We document the open-source hardware, firmware, and control protocols of the Generic Pan Tilt platform and reflect on design tradeoffs for accessibility, reliability, and expressivity. Finally, we outline a real-time analysis toolchain–motif detection, parallelism, and continuous energy/tension estimators–that emits OSC triggers for lighting, media, kinetic, and spatial-audio systems, enabling reactive shows beyond timecode. Together, these systems advance performable visualizations of human-improvised and AI-driven musical futures.

Design Rules for LLM-Generated Code: A RealWorld Case Study

Mon, 01 Sep 2025 00:00:00 GMT

Design Rules for LLM-Generated Code: A RealWorld Case Study Lawrence, Jennifer M. This thesis conducts a case study exploring the interaction between software design, extensibility, and LLM code generation. The central problem we investigate is whether LLMs violate software design principles in ways that introduce bugs and ultimately hinder extensibility. We examine several repositories belonging to the RealWorld collection, a project that demonstrates combinations of frameworks, database, and programming languages for building full stack web apps modeled on an existing social media application. We create a concept-based implementation of the RealWorld API. Concept Design defines software systems in terms of the abstract purposes and relationships of self-contained units of functionality. It enforces stringent design standards and aims to aid humans better understand complex software behavior. To test code extensibility, we develop three phases of new functionality to be added to the RealWorld API. Each phase is intended to mimic real-world software development, adding functionality that is commonly found in social media platforms while increasing nuance and complexity. The code for these extensions is generated by an AI agent, then reviewed by a human coder who classifies and fixes any bugs. In this study, we examine how LLMs interact with software paradigms like Concept Design, the kinds of design violations they produce, and whether these violations correlate with bugs that impede extensibility.

Cognify: An On-Device, AI-powered Learning Assistant

Mon, 01 Sep 2025 00:00:00 GMT

Cognify: An On-Device, AI-powered Learning Assistant Huang, Siyong Large Language Models (LLMs) have proven highly effective for a wide range of natural language processing tasks, but their size and compute requirements often restrict their use to powerful cloud-based infrastructures. In recent years, significant progress has been made in shrinking LLMs while maintaining performance levels comparable to much larger models. We are approaching the point where the capabilities of massive, multi-billion parameter models can be realistically replicated on consumer-grade devices. This thesis builds upon that foundation by developing an AI-powered note-taking application that runs entirely offline, using only the compute resources available on a personal laptop. The application is designed to listen to lectures alongside the student and provide support in real-time—through transcription, notes generation, and enabling context-aware search. Achieving this level of interactivity locally introduces challenges in reducing end-to-end latency, which this project addresses through both model-level optimizations and the design of efficient prompting and inference algorithms. A demo of the app can be found on Youtube.

Performance Analysis of the Apple AMX Matrix Accelerator

Mon, 01 Sep 2025 00:00:00 GMT

Performance Analysis of the Apple AMX Matrix Accelerator Zhou, Jonathan Apple Silicon integrates a dedicated Apple Matrix Coprocessor (AMX) that executes outer-product style computations with high throughput, but its public programming model remains largely hidden behind the Accelerate framework. This thesis turns AMX into a more predictable and practical target by combining (i) empirical throughput characterization, (ii) a case study on AMX specific matrix multiplication (GEMM) design, and (iii) an interpretable rule-based latency model that predicts cycle counts for short AMX instruction sequences. First, microbenchmarks quantify AMX load/store and compute limits across matrix and vector modes and data types. We analyze throughput in both GFLOPS and AMX instructions per cycle, and also observe output register based throughput limitations. Second, we develop an in-place GEMM that uses masked outer products and strategically overlapping tiles to avoid scratch buffers used by Accelerate, outperforming Accelerate while preserving simplicity. Third, we introduce a compact latency model that decomposes cycles into per-instruction BaseTime, symmetric SwitchLatency for instruction changes, and instruction FullLatency (data dependency) terms. Fitted with non-negative coordinate descent on length-2 loops and validated on length-3 sequences via a lightweight loop simulation, the model obtains reasonably high accuracy while remaining helpful for those trying to understand the architecture.

Weak Identification and Network Measurement Error in Peer Effects Estimation

Mon, 01 Sep 2025 00:00:00 GMT

Weak Identification and Network Measurement Error in Peer Effects Estimation Wang, William Wei The growing availability of social network data has enabled a surge of research on social interactions. In particular, peer effects, once considered unidentifiable, have now been shown to be identified given knowledge of the network structure. Despite this positive result, questions remain about the existence and nature of peer effects, due to concerns about identification strength and the reliability of network data. This work investigates two key threats to the estimation of peer effects: weak identification and network measurement error. We show that weak instrument problems arise in moderately dense networks due to rapid averaging, leading to slow convergence rates even when estimators remain consistent. On the measurement error side, we show that additive edge weight errors can be mitigated in such networks due to the same averaging phenomena, but the error remains a relevant threat to consistency in sparser networks. We further demonstrate that when both issues are present, the resulting estimators exhibit non-vanishing bias, suggesting that the combined effect of weak instruments and measurement error can be more severe than either problem in isolation. Overall, our results aim to clarify how these non-standard estimation challenges impact our ability to study peer effects using network data.

Seeing Beyond Limits with Physics-Informed Priors

Mon, 01 Sep 2025 00:00:00 GMT

Seeing Beyond Limits with Physics-Informed Priors Liu, Yang Conventional imaging systems are limited by dimensionality and visibility: standard sensors capture only two-dimensional data, while light diffuses or scatters across surfaces and through complex media. This dissertation reformulates imaging as an interplay of optical encoding and neural decoding. It models forward physical processes and iteratively refines them using deep denoisers. By embedding physics-informed priors into this optimization, it aims to surpass conventional limits in dimensionality and visibility. First, I develop Privacy Dual Imaging using an ambient light sensor. This approach tackles both dimensionality and visibility challenges when imaging with a single-point, non-imaging component on smart devices. Inspired by 1984’s “Big Brother” telescreen, I demonstrate how subtle light intensity fluctuations can reveal unseen image information; however, the goal is to highlight privacy concerns, not exploit them. It addresses two visibility limits—pixel-less and lens-less imaging—by using the screen as a spatial modulator and exploiting involuntary motion to create a virtual pinhole effect. A quantized, physics-informed prior improves reconstruction from heavily quantized sensor measurements. Second, I propose Snapshot Compressive Imaging (SCI) augmented with deep plug-and-play physics-informed priors to overcome the dimensionality limit of 2D sensors. SCI compressively encodes multiple temporal, spectral, or angular frames into a single measurement. A deep plug-and-play prior algorithm introduces high-dimensional priors learned from images and videos into the iterative reconstruction process, improving fidelity, speed, and flexibility. Experiments show notable gains in reconstruction quality and efficiency across different SCI datasets, including largeformat 4K UHD scenarios. Third, I introduce Rank-Reduced physics-informed priors, showing that large pretrained AI models—especially diffusion models—can act as general visual priors across both dimensionality and visibility challenges. A relax-then-tighten strategy handles ill-conditioning by applying truncated singular value decomposition to reduce rank deficiencies, followed by a Stable Diffusion refiner (SDEdit) plug-and-play prior that constrains reconstructions to valid image spaces. Simulations and passive non-line-of-sight imaging experiments verify the approach’s stability and effectiveness. Physics-informed priors promise to extend the boundaries of imaging, enabling us to see beyond current dimensionality and visibility limits and to unlock new applications from macro-scale to micro-scale observations.

Optimizing Large Language Models from a Data SystemsPerspective

Mon, 01 Sep 2025 00:00:00 GMT

Optimizing Large Language Models from a Data SystemsPerspective Chen, Peter Baile Strong retrieval and reasoning capabilities are essential for large language models (LLMs) to effectively handle a broad spectrum of downstream tasks, such as open-domain question answering and solving math or science problems. While current LLM-based frameworks achieve strong performance on complex retrieval and reasoning tasks, they do so at a high computational cost. Additionally, they often lack structured, systematic problem-solving strategies, leading to unexpected failures. In particular, these models typically operate in an iterative, online, and isolated fashion—failing to exploit relationships across data sources, opportunities for offline computation, and the benefits of reusability—resulting in less-than-optimal outcomes. In contrast, traditional data management systems are engineered for both efficiency and accuracy, with careful coordination across all stages of the query pipeline. Inspired by these principles, this work proposes novel approaches to improve LLMbased retrieval and reasoning by incorporating optimization techniques from data systems. Our evaluation across a range of knowledge- and reasoning-intensive datasets demonstrates significant gains in both accuracy and computational efficiency.

Foundational Abstractions for Quantum Programming

Mon, 01 Sep 2025 00:00:00 GMT

Foundational Abstractions for Quantum Programming Yuan, Charles Bringing the promise of quantum computation into reality requires not only building a quantum computer but also correctly programming it to run a quantum algorithm. To obtain asymptotic advantage over classical algorithms for applications including simulation, search, and optimization, quantum algorithms rely on the ability of data in quantum superposition to exhibit phenomena such as interference and entanglement. In turn, an implementation of the algorithm as a program must correctly orchestrate these phenomena in the states of qubits. Otherwise, it would yield incorrect outputs or lose quantum computational advantage. Given a quantum algorithm, what are the challenges and costs of realizing it as a program that can run on a physical quantum computer? In this thesis, I answer this question by showing how the basic abstractions of programming upon which many quantum algorithms rely – such as data structures and control flow – can fail to work correctly or efficiently on a quantum computer. I then demonstrate how we can leverage insights from research in programming languages to re-invent the software stack – including abstractions, libraries, and compilers – to meet the demands of quantum algorithms. This approach holds out a promise of expressive and efficient tools to program a quantum computer and thereby practically realize its computational advantage.

Fabrication of Superconducting Reflectionless Filters for Quantum Microwave Circuits

Mon, 01 Sep 2025 00:00:00 GMT

Fabrication of Superconducting Reflectionless Filters for Quantum Microwave Circuits Bui, Eric The performance and scalability of superconducting quantum circuits depends critically on the microwave environment. Minimizing signal reflections and suppressing thermal noise are essential for achieving high-fidelity readout and preserving qubit coherence. A significant challenge arises from the use of conventional cryogenic components such as isolators and circulators, which exhibit nonideal out-of-band reflection characteristics. Reflections degrade impedance matching and limit the performance of broadband quantum limited amplifiers. Superconducting implementations of reflectionless microwave filters offer a promising solution to mitigate these issues. The focus of this work is the fabrication and cryogenic characterization of reflectionless filters compatible with superconducting qubit fabrication flows. Devices were implemented on high resistivity silicon substrates using aluminum ground planes, integrated nichrome resistors, and crossovers formed with SiO2 interlayer dielectric. Cryogenic measurements at 20 mK demonstrate high return loss, confirming the viability of these filters for co-fabrication with traveling-wave parametric amplifiers (TWPAs) and circuit quantum electrodynamics (cQED) architectures. The filters exhibit low insertion loss in the passband to maintain quantum measurement efficiency and provide broadband reflection suppression across frequencies relevant to superconducting qubits, offering a scalable way to manage microwave noise in superconducting quantum processors.

CONDOR: Clinical Ontology-aware Networked Data Organization and Retrieval

Mon, 01 Sep 2025 00:00:00 GMT

CONDOR: Clinical Ontology-aware Networked Data Organization and Retrieval Dongo Aguirre, Gyalpo Melchisedeck Until now, state-of-the-art research into AI-driven clinical workflows has been confined to proprietary, closed-source systems from vendors like Epic and Oracle, or private experiments like Stanford’s ChatEHR, creating a critical barrier to academic innovation. This thesis introduces CONDOR, the first fully open-source and replicable research environment designed to simulate an agentic, conversational AI interacting with a high-fidelity Electronic Health Record (EHR). By integrating an open-source, FHIR-native EHR (Medplum) with a complex, realistic public clinical dataset (MIMIC-IV FHIR), CONDOR provides a foundational testbed that has been previously unavailable to the research community. The framework’s primary contribution is a novel alignment and evaluation methodology that adapts the principles of SelfCite to the clinical domain. We propose a ‘ClinicalConfidence‘ score to quantify the trustworthiness of generated statements and programmatically generate a high-quality preference dataset for alignment using Simple Preference Optimization (SimPO). We compare a standard vector-based Retrieval-Augmented Generation (RAG) baseline against a more advanced GraphRAG architecture that leverages a two-tiered knowledge graph of patient data and medical ontologies. Our results demonstrate that the full CONDOR system, combining GraphRAG with SimPO alignment, significantly improves citation quality and verifiability, establishing a new open-source benchmark for the development of safe and reliable clinical AI.

Multi-Stage LLM Reasoning for Automated Detection and Classification of High-Impact Misinformation

Mon, 01 Sep 2025 00:00:00 GMT

Multi-Stage LLM Reasoning for Automated Detection and Classification of High-Impact Misinformation Nair, Anushka Manchanda As of 2025, social platforms have become a primary news source, magnifying the reach of misleading content [1]. Exposure to misinformation has been linked to shifts in public attitudes and behavior, including vaccine uptake [2] and voting behaviors [3]. However, current misinformation detection approaches can often focus on a narrow definition of misinformation: factual claims that can be clearly judged as true or false. However, recent research suggests the problem lies elsewhere: overt falsehoods (“vaccines contain microchips”) can carry little harm, while technically accurate but decontextualized narratives can be more influential. Allen et al. (2024) [4] found that factually accurate ”vaccine-skeptical” content had a much greater impact on vaccine hesitancy than misinformation flagged by fact-checkers. These narratives can work by omitting information, misleading framing, or cherry-picked evidence, forms of manipulation that can elude traditional fact-checking. Though professional fact-checkers are often able to recognize these tactics and the broader context of information, they cannot keep pace with the volume of online content. This thesis designs a Large Language Model (LLM) based pipeline meant to partner with, rather than replace, human fact checkers. The system decomposes content into its explicit and implicit claims, rhetorical tactics, and the “missing context” questions it raises; retrieves evidence from fact-check databases and reliable sources; and synthesizes grounded explanations while assigning calibrated harm scores to guide triage. Evaluated on fact-checked tweets, the pipeline matched expert judgments in 92.6% of cases where experts agreed, and flagged for review posts where experts disagreed, a gray zone requiring human judgment. The system’s explanations ranked higher than crowdsourced Community Notes in helpfulness, clarity, and trustworthiness when assessed by an LLM, and harm evaluations aligned with human reviewers in 87.5% of cases, enabling prioritization of content with greatest potential impact. Despite constraints of sample size and processing latency, the results demonstrate the feasibility of a human–AI workflow that treats disagreement as a signal and directs scarce attention towards high-impact misinformation that current automated systems can miss.

Learning Simple Chemical Heuristics to Model and Discover Materials

Mon, 01 Sep 2025 00:00:00 GMT

Learning Simple Chemical Heuristics to Model and Discover Materials Ma, Andrew Computational approaches have long played an important role in the field of materials science, driving both the scientific study of materials’ fundamental properties and the design of materials for technological applications. Currently, mainstream methods in computational materials science typically rely on either first-principles calculations or deep learning models. In this thesis, we take a different direction by developing remarkably simple data-driven models for predicting fundamental properties of materials, including electronic topology, metallicity, and band gap. These models take the form of highly interpretable chemical heuristics. A key finding of this work is the surprising result that electronic topology diagnosis – often regarded as a highly complex task – can, in fact, be performed heuristically using a simple and intuitive model. We further integrate this model into a workflow for discovering new topological materials. Altogether, this work revisits the classic idea of chemical heuristics through a modern data-driven lens, shedding new light on fundamental problems in materials science.

Spatial Emission and Polarization Control in Integrated Photonics for Optical-Trapping and Trapped-Ion Systems

Mon, 01 Sep 2025 00:00:00 GMT

Spatial Emission and Polarization Control in Integrated Photonics for Optical-Trapping and Trapped-Ion Systems Sneh, Tal Recent advances in silicon photonics have yielded impressive results in fields including biophotonic optical tweezers and trapped-ion quantum systems. However, the majority of these demonstrations, while offering advantages in size, cost, and dense integration, lag behind their bulk-optic counterparts, limited by a lack of critical advanced functionality such as spatial control of light in the near field or polarization control at visible wavelengths. This thesis addresses this gap by designing and experimentally demonstrating the first, to the best of our knowledge, cell experiments using single-beam integrated optical tweezers, chip-based 3D printers, and integrated polarization rotators and splitters at blue wavelengths. First, we demonstrate optical trapping and tweezing of microspheres using a nearfield-focusing integrated optical phased array, at a standoff distance over two orders of magnitude larger than prior integrated demonstrations. We then use this system to perform the first cell experiments using single-beam integrated optical tweezers. Second, we use a tunable integrated optical phased array operating at red wavelengths to print designs in a visible-light-curing resin, demonstrating the first chip-based 3D printer. Third, we design and experimentally demonstrate the first integrated polarization rotators and splitters operating at blue wavelengths, enabling polarization control on chip for sophisticated integrated manipulation of trapped-ion and neutral-atom quantum systems. Finally, we develop key polarization-diverse integrated-photonics devices and utilize them to implement a variety of integrated-photonics-based polarization-gradient-cooling systems, culminating in the first demonstration of polarization-gradient cooling of a trapped ion by an integrated-photonics-based system.

Language Modeling from Visually Grounded Speech

Mon, 01 Sep 2025 00:00:00 GMT

Language Modeling from Visually Grounded Speech Lai, Cheng-I Jeff Recent advancements in spoken language processing have significantly reduced automatic speech recognition (ASR) error rates, driven by large-scale supervised training on paired speech–text data and, more recently, self-supervised pre-training on unpaired speech and audio. These methods have facilitated robust transfer learning across diverse speech and audio tasks. However, fully leveraging multimodal inputs, particularly visual context, remains underexplored. This thesis addresses this gap by developing novel language modeling techniques directly from visually grounded speech. We first introduce the Audio-Visual Neural Syntax Learner (AV-NSL), an unsupervised parser that recovers constituency trees directly from raw speech paired with images, demonstrating how visual context effectively bootstraps grammar induction without textual supervision. Next, we investigate Audio-Visual Word Discovery for Speech Translation, using the Fisher Spanish–English corpus to train a series of speech-to-speech translation models based on pseudo-word units discovered via audio-visual grounding. This study highlights that simplistic acoustic tokens and limited training data degrade re-synthesis and translation quality, underscoring two crucial missing ingredients: richer semantic tokens and large-scale training. Guided by these insights, we present Audio-Visual Gemma (AV-Gemma), a family of multimodal foundation models that condition jointly on images and learned semantic speech tokens. At scale, AV-Gemma generates visually coherent spoken captions and transfers robustly to tasks such as video-to-speech generation and spoken visual question answering, significantly advancing multimodal spoken-language processing.

ALPACA: An Algorithmic Pipeline for Automated Contour Annotation of Carnatic Music: A Dynamic Programming Framework for Pitch Segmentation and Note Transcription

Mon, 01 Sep 2025 00:00:00 GMT

ALPACA: An Algorithmic Pipeline for Automated Contour Annotation of Carnatic Music: A Dynamic Programming Framework for Pitch Segmentation and Note Transcription Parthasarathi, Sruthi In recent years, a wide range of computational techniques have been developed to extract information from recorded performances of Western music. However, these methods often achieve limited success when applied to non-Western musical traditions. Carnatic music, in particular, poses unique challenges due to the absence of a standardized notation system and the lack of a consistent mapping between frequency bands and note categories. This project introduces a dynamic programming–based transcription framework, incorporating novel methods for label estimation, contour segmentation, and related subtasks, and establishes the foundations for end-to-end automatic transcription of this art form.

Modeling Diverse Treatment Policies from Observational Health Data

Mon, 01 Sep 2025 00:00:00 GMT

Modeling Diverse Treatment Policies from Observational Health Data Ejilemele, Abe Learning policies for real world tasks often requires modeling human behavior, especially in domains like healthcare and driving. In these settings, skills are learned from expert human demonstrations, but such data are typically multimodal, violating the common single expert assumption. We study sequential clinical treatment decision making in the offline imitation learning setting, where environment interaction is prohibited, reflecting the challenges of experimentation in safety critical domains. Existing methods for multi expert offline imitation learning often restrict the latent space, underspecify its structure, or omit objective terms that prevent latent collapse and encourage behavior discovery. We propose a fully offline approach that addresses these shortcomings and improves learning from multi expert demonstrations through modifications to the formulation of the latent approximate posterior and the model architecture. We suggest that our method is more robust to real world settings where the true number of demonstrators may not be known. We also incorporate an occupancy matching term into our objective that injects awareness of the rollout distribution over trajectories into our behavior cloning objective. We evaluate our method against baselines on both simulated multi expert demonstrations from an extended S-CVSim and real world demonstrations from MIMIC. Our approach achieves consistently higher next step action prediction and behavior discovery performance. While ground truth expert policies are unavailable for MIMIC, visual analysis shows our method uncovers clinically meaningful variations in expert strategies, reflecting treatment population diversity.

Scalable Assembly of General Objects

Mon, 01 Sep 2025 00:00:00 GMT

Scalable Assembly of General Objects Tian, Yunsheng In this thesis, I present a scalable system towards fully automated and flexible robotic assembly that generalizes over diverse geometries and complex structures. Most real-world objects are assemblies composed of multiple parts. Assembly presents significant challenges for robots to execute long-horizon, contact-rich manipulation with both reliability and generalization. However, most manufacturing facilities today still rely heavily on manually programmed assembly lines, which require significant labor, time, and setup costs yet offer no flexibility to object variations. My proposed system synergizes global multi-step planning with local reactive learning-based control to enable generalizable and precise assembly. Such an integrated paradigm effectively leverages the best of both worlds, accomplishing results that neither planning nor learning could achieve alone. For planning, I leverage guidance from physical simulation and learned feasibility networks to efficiently search for part sequences, precise motions, and stable grasps for dual-arm robots over long horizons. For learningbased control, I train robust policies via reinforcement learning for submillimeter-level insertion across different part geometries, assembly paths, and grasp poses. I introduce and open-source the largest-scale assembly dataset to date and demonstrate my system’s generalization on thousands of simulated assemblies as well as through end-to-end real robot experiments. By integrating planning and learning, I showcase the first system to achieve complete and generalizable real-world multi-part assembly without domain knowledge or human demonstrations. Although the system plans and learns purely in simulation, it transfers zero-shot to the real world and achieves 80% successful steps. Finally, I will share insights that further scale up robotic assembly and opportunities to extend to general manipulation, and discuss future directions to equip general-purpose robots with multi-step, precise manipulation capabilities.

Towards a Modular Superconducting Quantum Processor using Chiral Waveguide Quantum Electrodynamics

Mon, 01 Sep 2025 00:00:00 GMT

Towards a Modular Superconducting Quantum Processor using Chiral Waveguide Quantum Electrodynamics Yankelevich, Beatriz As the field of superconducting quantum computing advances, networking qubits within a single system becomes essential for building modular processors. Modularity allows the system to circumvent scalability constraints and enable architectures and computational schemes that exploit non-local connectivity to enhance processing capabilities. This work proposes non-local entanglement generation methods based on the theory of chiral quantum waveguide dynamics, which is the quantum-optical framework that describes systems of atoms coupled non-reciprocally to a continuum of modes. We leverage these effects to design a chiral communication module composed of multiple superconducting qubits, capable of both directional single photon routing and the realization of chiral, driven-dissipative entanglement protocols.

Fine-tuning Boltz for Antibody-Antigen Binding Prediction

Mon, 01 Sep 2025 00:00:00 GMT

Fine-tuning Boltz for Antibody-Antigen Binding Prediction Kim, Ji Won Accurate prediction of antibody-antigen binding is a central challenge in computational immunology. Its direct implication for therapeutic antibody design and vaccine development has made it one of the most rapidly growing fields. Recent advances in protein language models and structure prediction have provided new tools for modeling, yet these approaches often fall short in capturing the fine-grained features that drive binding specificity in antibody and antigens. This thesis evaluates multiple strategies for improving predictive performance. First, we investigate a custom multiple sequence alignment (MSA) experiment. Standard Boltz-2 training relies on MSAs from broad protein databases, which capture global diversity but under-represent lineage-specific constraints. To address this, we constructed antibody-specific MSAs to test whether restricting the search space to antibody repertoires improves model learning. Unfortunately, gains in downstream binding prediction were limited, suggesting that further work needs to be done in training models for specific databases in the first place. Our second line of investigation focused on fine-tuning Boltz-2, a generative structural foundation model, using curated antibody–antigen data. By leveraging Boltz-2’s internal sequence embeddings, we trained a predictive model for binding affinity. This approach yielded stronger ROC performance compared to baseline models, achieving a validation AUROC of 0.645, demonstrating the advantages of structural generative priors for antibody–antigen binding prediction.

Deterministic Circuit Range Avoidance is (Likely) Intractable

Mon, 01 Sep 2025 00:00:00 GMT

Deterministic Circuit Range Avoidance is (Likely) Intractable Ilango, Rahul Circuit Range Avoidance (denoted Avoid) is a computational problem where, given a Boolean circuit with more output bits than input bits, one must output a string outside of the range of the circuit. A simple counting argument implies that such a string must always exist and also guarantees that outputting a uniformly random string is correct with good probability. A natural question is whether this can be derandomized: does there exist an efficient deterministic algorithm for Avoid? We give the first evidence that deterministically solving Avoid is intractable. We show that there is no polynomial-time algorithm for Avoid under plausible assumptions in complexity theory and cryptography. Specifically, our assumptions are that NP ≠ coNP and that subexponentially-secure indistinguishability obfuscation exists.

Learning to Tackle Task Variations in Control - A Transportation Context

Mon, 01 Sep 2025 00:00:00 GMT

Learning to Tackle Task Variations in Control - A Transportation Context Jayawardana, Vindula Muthushan Real-world control tasks are messy and often exhibit task variations. Practical solutions to these problems must exhibit generalization across task variations. For example, in the task of controlling traffic signals, control strategies must adapt to different intersection topologies (the variations), each with distinct dynamics. In this thesis, we consider the challenge of coping with task variations in the context of transportation problems, specifically in roadway interventions where many such variations are both common and imperative to handle. We develop machine learning techniques to address three key challenges: 1) quantify the impact of task variations in control, 2) model them to align with the real world, and 3) optimize in the presence of them. To this end, we begin with a large-scale case study of cooperative eco-driving and illustrate how explicitly modeling task variations can surface otherwise overlooked insights. Building on this, we argue for the necessity of formally incorporating task variations into problem specifications, emphasizing that task underspecification due to loosely defined task variations can severely impair decision-making. We then introduce a contextual reinforcement learning algorithm capable of leveraging the structure of task variations to generalize effectively in cooperative eco-driving with autonomous vehicles. We also present IntersectionZoo, a benchmark designed to promote the development of learning algorithms that generalize by exploiting task variation structures, thus standardizing progress in the field. Last, we explore task variation modeling through a generative modeling lens, using human driver behavior modeling as a case study. Overall, this thesis lays the groundwork for robust control methods by leveraging machine learning to tackle task variations, specifically in roadway intervention designs.

An Empirical Evaluation of LLMs for the Assessment of Subjective Qualities

Mon, 01 Sep 2025 00:00:00 GMT

An Empirical Evaluation of LLMs for the Assessment of Subjective Qualities Ranade, Esha Large Language Models (LLMs) have achieved remarkable success in natural language processing tasks and are increasingly being used for language generation. Significant advancements in this field have unlocked capabilities that enable their adoption in sophisticated roles, including acting as evaluators or "judges" of text for various attributes such as factuality, relevance, fluency, and reasoning quality. However, their understanding and ability to assess subjective attributes, such as the level of formality in a piece of writing, and produce content matching these subjective attributes remains unclear and underexplored. This research develops a methodology to study how LLMs evaluate subjective attributes. It has three primary contributions: (i) a reproducible user study to generate human-annotated labels for different attributes, (ii) an analysis of the extent to which different LLMs provide subjective labels aligned with human annotators, and (iii) an analysis of the extent to which LLMs generate content aligned with specified intended subjective labels, relative to humans. The user study and the analyses have been conducted both with and without a reference scale. The scale itself, the survey design, and the evaluation questions have all undergone multiple rounds of iteration informed by study tester feedback to improve clarity, consistency, and reliability for the final study. Comparisons between human-generated ratings and LLM-generated ratings for both human-generated content and LLM-generated content reveal the extent to which LLMs align with human judgment, providing insights into their capabilities and limitations. While humans typically do better in their roles, LLMs are able to attain reliably high levels of success in producing and judging text, despite tending to err on the more-formal side. Both groups’ performance increases significantly with the aid of a formalized reference scale. Across the suite of models tested, OpenAI’s GPT family leads overall performance, with Anthropic’s Claude and Meta’s LLaMA series showing notable strengths in specific formality ranges. Although this work focuses on the formality attribute of text, the methodology developed can be used to evaluate other subjective qualities of text, such as conciseness, usefulness, or persuasiveness. Ultimately, these findings may guide future efforts to fine-tune LLMs to produce text that more precisely matches the desired stylistic or ethical standards.

Accelerating Burst Parallelism of SigmaOS processes with CRIU

Mon, 01 Sep 2025 00:00:00 GMT

Accelerating Burst Parallelism of SigmaOS processes with CRIU Tang, Frederick σOS is a multi-tenant cloud operating system designed to integrate the agility of serverless environments with the interactivity of microservices. A goal of achieving this integration is the ability to start new instances of server processes quickly. However, σOS only handles σcontainer initialization, and does not assist with runtime and app initialization costs. One approach to overcome this challenge is to checkpoint processes using Checkpoint/Restore in Userspace (CRIU). CRIU is a linux toolset which can start new server instances by restoring them from a saved checkpointed state, avoiding the full cost of reinitialization and setup. This thesis introduces σCRIU, which adapts CRIU for burst-parallel spawning of microservices in σOS. σCRIU implements a number of optimizations: compressing checkpointed proc metadata to reduce network communication costs, implementing demand-paging using a lazy page service, and caching kernel metadatadata to reduce CRIU’s restore operation latency. These optimizations allow σCRIU to start new microservices on remote machines quickly while still making use of CRIU’s existing proven checkpoint and restore technology.

Modern methods for causal inference and missing data

Mon, 01 Sep 2025 00:00:00 GMT

Modern methods for causal inference and missing data Xia, Eric The proliferation of data-driven approaches in a wide array of settings is one of the defining characteristic of the modern era. With this rise, there has been much focus on using data to answer causal questions, e.g. whether A causes a change in B. Furthermore aspects of data collection has given rise to datasets that are often quite messy, sometimes missing important entries. These are both problems that are incredibly relevant to practitioners in a variety of disciplines, including policy-makers looking to make critical decisions that can influence lives of many. On the surface these problems seem quite distinct, yet the literature has highlighted deep connections between these two settings. Indeed, many methods for addressing one question can often be repurposed to address the other. These two settings are quite classical and approaches to address the are still quite so, but there has been great interest recently to develop techniques and algorithms to address them that harness modern developments in statistics and machine learning. This thesis contributes to the literature by providing new methods as well as novel understandings of existing ones.

Visually Accurate Database-Enabled Reconstructions of Scenes (VADERS)

Mon, 01 Sep 2025 00:00:00 GMT

Visually Accurate Database-Enabled Reconstructions of Scenes (VADERS) Gosalia, Mehek This work introduces a novel pipeline for scene reconstruction that jointly prioritizes semantic accuracy and visual fidelity, addressing a gap in current approaches. Prior pipelines often emphasize either semantic analysis or photorealistic rendering, but rarely both. This method combines scene analysis, segmentation, and retexturing to yield reconstructions that preserve structural semantics, while convincingly reflecting the visual qualities of the original image. The motivation lies in the limitations of existing systems. Existing databaseassisted approaches depend on proprietary datasets that restrict stylistic diversity or using in-the-wild assets. This constrains expressiveness and often produces results that are visually misaligned. Conversely, pipelines optimized for visual realism neglect semantic correctness, generating outputs that may appear plausible but lack categorical or structural grounding. Our framework addresses this by first enforcing semantic accuracy via selecting database assets, then editing those assets to be stylistically faithful to the reference, producing reconstructions that are both interpretable and expressive. We begin with database-assisted scene analysis, using an open-source asset database containing chairs, lamps, sofas, tables, and benches. Input images are depth-mapped, segmented, and parsed into object masks, which are matched to database assets based on semantic labels and visual correspondence. Each asset is broken into semantic segments and rescaled per-component using vision-language model predictions to match the reference object better. Finally the asset is retextured based on the image mask of the reference object in the input image. Evaluation on six diverse scenes—both photographs and artworks—shows the pipeline produces semantically grounded, visually accurate reconstructions under non-research conditions. Future work will focus on expanding the asset database, reducing reliance on proprietary texturing, and releasing an open-source implementation to broaden accessibility.

Designing Planar Silicon Solar Cells for Singlet Fission Sensitization

Mon, 01 Sep 2025 00:00:00 GMT

Designing Planar Silicon Solar Cells for Singlet Fission Sensitization Wang, Janet Z. Singlet fission (SF)-sensitized silicon (Si) solar cells offer a path towards surpassing the Shockley-Queisser efficiency limit for single-junction solar cells. However, realizing efficient charge transfer from the SF material to Si remains a significant challenge that requires careful interface engineering. Prior work showed that Si microwire cells sensitized with tetracene (Tc) and a zinc phthalocyanine (ZnPc) donor layer can boost photocurrent and external quantum efficiency (EQE). Planar devices are simpler to fabricate than microwire devices and reproduce the planar geometry of optical test samples to connect studies of the interface to device performance. This thesis integrates modeling and experimental approaches to guide the design of planar SF-sensitized Si solar cells. We developed a fabrication process for planar cells comparing varied oxide passivation layer growth conditions and surface treatments, Si(100) versus Si(111) orientation, and junctions formed by diffusion doping versus ion implantation. Complementary surface photovoltage (SPV) measurements on matching optical stacks show evidence of an illumination-induced transient positive charge density at the Tc/ZnPc/oxide/Si interface, consistent with increased field effect passivation. We find that SPV responses on AlOx/n-Si are dominated by substrate band bending; consequently, SiOx is the preferred passivation to suppress the background and isolate the SPV signals driven by the organics. A drift–diffusion model shows that the diffusion doping (exponential) emitters reduce surface recombination rates compared to ion implantation (Gaussian) emitters. We also show that a positive fixed charge density at the surface enhances short wavelength EQE, with the effect strongest for Gaussian emitters. Together, these results provide practical design rules for planar SF-sensitized Si cells and the study of charge transfer at organic-Si interfaces.

Microsecond Time Synchronization for Computing Fiber Networks

Mon, 01 Sep 2025 00:00:00 GMT

Microsecond Time Synchronization for Computing Fiber Networks Li, Jenny Y. We present a microsecond-accurate time synchronization method and time localization system for a sensor network of spatially-separated, low-power Bluetooth nodes, with the goal of integrating this system into thermally-drawn computing fibers. Each node consists of an nRF54L15 SoC paired with an ICS-43434 digital I2S microphone, enabling synchronized audio data collection. Our design leverages Bluetooth LE connection events to synchronize local clocks with sub-10 µs accuracy across a multi-peripheral topology; we trigger precise, CPU-independent hardware events to timestamp audio samples. We demonstrate that timestamped I2S data stored in external SPI flash can be correlated across devices to extract TDoA measurements for localizing sound sources. Cross-correlation techniques allow us to estimate direction and position, with localization errors reduced from 4.17 m to 0.39 m through clock synchronization. This prototype provides a roadmap for embedding synchronized sensing and computation within fibers and smart textiles, with implications for on-body audio perception and distributed sensing in flexible electronics.

From String to Structure: Graph Threading for Physical Assembly

Mon, 01 Sep 2025 00:00:00 GMT

From String to Structure: Graph Threading for Physical Assembly Lin, Rebecca Y. E. Many artistic and engineering applications—from beadwork to deployable structures—create intricate, and sometimes dynamic, designs by threading cord through tubular components. We model the underlying design challenge—threading tubes so that they achieve a target connectivity when the string is pulled taut—as graph threading. In this formulation, tubes and their junctions correspond to edges and vertices of a graph, and the goal is to find a closed walk that induces a connected graph at every vertex while avoiding U-turns. We study two optimization objectives motivated by fabrication and deployment: minimizing length to reduce material cost and assembly time, and minimizing turn to reduce frictional resistance during deployment. For the length metric, we present a polynomial-time algorithm via reduction to minimum-weight perfect matching, prove tight worst-case bounds on optimal threadings, and identify special cases with faster algorithms. For the turn metric, we characterize the complexity landscape, proving NP-hardness for graphs of maximum degree 4, tractability for degree 3, and giving exact and approximation algorithms for restricted variants, including rectangular grid graphs. Finally, we turn from theory to fabrication, proposing multi-configuration threading—a new approach for achieving multiple predetermined configurations within a single system. As in earlier chapters, framing the problem in graph-theoretical terms provides access to powerful problem-solving techniques, guiding both algorithmic analysis and physical design.

Steering Vision at Scale: From the Model Weights to Training Data

Mon, 01 Sep 2025 00:00:00 GMT

Steering Vision at Scale: From the Model Weights to Training Data Materzyńska, Joanna We study the interpretability and controllability of multimodal and generative models, with a particular focus on text–image representation models and text-to-image diffusion systems. We begin by addressing limitations in CLIP’s multimodal embeddings, specifically the entanglement between visual and textual concepts within images. We demonstrate the consequences of this entanglement in both generative and discriminative tasks, and introduce a method for disentangling visual and textual representations. We showcase the utility of these disentangled embeddings in typographic attack resistance, improved image generation, and robust out-of-domain OCR detection. Building on this foundation, we explore methods to enhance the controllability of diffusion models. First, we tackle the challenge of unwanted concept generation. We introduce a technique to remove specific visual concepts using only their names, leveraging negative prompts and guidance to suppress target content without modifying training data or requiring model retraining. This approach enhances ethical alignment and enables greater user control in generative systems. We then turn to the complementary problem: incorporating new concepts. We present a few-shot motion customization technique for video generation models, which transfers motion patterns from a small set of examples to novel subjects. This method maintains the generalization capabilities of the base model while enabling consistent, subject-agnostic animation that preserves both identity and temporal coherence. To improve the fine-grained control of visual outputs, we propose a method for continuous manipulation of image attributes. This framework introduces smooth, intuitive controls, that allow for dynamic, continuous steering of generated images. Unlike prompt engineering or token-level interventions, our approach offers real-time adjustment without sacrificing output realism. Finally, we examine whether artistic styles in diffusion models require large-scale pretraining or can be learned in a lightweight, post-training manner. To this end, we train a base model on art-free data and introduce a compact adapter method that learns stylistic concepts from a small set of exemplar artworks. Our findings suggest that artistic domains can be integrated efficiently and ethically, without reliance on web-scale scraped datasets.

A Data Attribution-Based Approach to Model Diagnosis in LC-MS/MS Structure Prediction

Mon, 01 Sep 2025 00:00:00 GMT

A Data Attribution-Based Approach to Model Diagnosis in LC-MS/MS Structure Prediction Khoo, Ling Min Serena Elucidating the structure of small molecules from complex mixtures using liquid chromatography tandem mass spectrometry (LC-MS/MS) is a challenging task with far-reaching implications in many areas such as drug discovery, environmental science and metabolism research. Yet, despite its importance and significant efforts to develop machine learning (ML) models for the task of elucidating the molecular structures of unknown compounds from LC-MS/MS spectra, the performance of these ML-based models remains limited. As a result, the performance of current ML-based models has been reported as insufficient for practical applications, thereby warranting a deeper investigation into their limitations to advance ML-based molecular structure elucidation from LC-MS/MS and enable their utility in real-world settings. Here, we leverage data attribution methods to systematically identify and validate hypotheses about the sources of generalization challenges that hinder current model performance. Our goal is to automatically uncover insights into the failure modes of existing ML models for LC-MS/MS, thereby laying the foundation for developing more robust and accurate models.

Modeling Dynamic Objects in Scenes with Generative Particle Systems

Mon, 01 Sep 2025 00:00:00 GMT

Modeling Dynamic Objects in Scenes with Generative Particle Systems Li, Eric Humans readily interpret the motion of deformable and rigid bodies, even when encountering unfamiliar objects with minimal shape or texture cues. In such cases, motion serves as a critical signal for recognition and understanding. Inspired by this ability, we propose a generative model that represents 3D matter as small Gaussians (“particles”) drawn from clusters capturing groups of coherently moving matter. We develop an e!cient inference algorithm based on parallelized block Gibbs sampling to recover stable particle motion and rigid groupings. Our model provides a tractable, object-centric generalization of as-rigidas-possible (ARAP) regularizers used in motion tracking. To assess alignment with human perceptual judgments, we test our approach on random dot kinematograms—sparse motion displays in which dot trajectories convey latent object structure, often used to probe visual understanding of motion and grouping. In this setting, our approach captures human-like responses, including graded patterns of uncertainty across ambiguous conditions. Applied to naturalistic RGB videos, it infers dense particle representations that track object motion and deformation over time. These results demonstrate that our model enables persistent latent scene structure suitable for object-level reasoning.

The Arm Qubit for Faster, Higher Fidelity Readout and Gates

Mon, 01 Sep 2025 00:00:00 GMT

The Arm Qubit for Faster, Higher Fidelity Readout and Gates Kline, Jeremy B. Currently, superconducting qubit processors are bottlenecked by errors during two-qubit gates, readout, and idle time. All three error contributions could be reduced if we improved the speed of operations (without introducing additional leakage errors) compared to the qubit lifetime. Readout and two-qubit gates are multimode interactions and therefore are limited by the coupling strength between the modes. In this thesis, we introduce a two-mode superconducting qubit which uses one mode to facilitate strong coupling to other modes of the quantum processor and one mode to store data with high coherence. Simulations show that this architecture could enable order-of-magnitude reductions in error during readout and two-qubit gates.

Clustering Algorithms for Component Placement in Printed Circuit Boards

Mon, 01 Sep 2025 00:00:00 GMT

Clustering Algorithms for Component Placement in Printed Circuit Boards Petrusenko, Vlada In 2024, approximately 12 billion printed circuit boards (PCBs) were manufactured globally [1], with the trend increasing gradually, and the majority of PCB layouts still being completed manually. The manual design process amounts to millions of hours of tedium that can be eased with automation. One of the biggest challenges is that the complex Printed Circuit Board designs typically have hundreds, sometimes thousands of components and even more net connections between them. This makes both manual and automated placement very time-consuming. As a way to improve placement performance, in this thesis, we constructed a custom weighted undirected graph representation of components and nets for any board that would encode physical and electrical constraints. Additionally, we integrated the Louvain and Leiden clustering algorithms for component clustering in PCB placement. We also showed comparative metrics with the spectral clustering algorithm applied to unweighted graph representations, which is the prior state of this project, but it has no knowledge of electrical and physical constraints associated with PCB designs and would thus produce results that require more manual correction. This new clustering approach was able to generate more optimal clustering and reduced average runtime by 51.05%, decreased estimated length of routing by 7.72%, and improved component association score by 12.8%.

How Data Drives ML Models Performance

Mon, 01 Sep 2025 00:00:00 GMT

How Data Drives ML Models Performance Khaddaj, Alaa Data has been been playing an increasingly more important role in the machine learning (ML) pipeline. This thesis deepens the understanding of the effect of the data on model performance and reliability. First, we study how choice of training data affects model performance. We consider a transfer learning setting and present a framework for selecting from a large pool of data a pretraining subset that improves model performance on downstream tasks. Our approach, however, requires training multiple target models which becomes prohibitively expensive at large-scale. To that end, we explore using smaller—and cheaper—proxy models to approximate large model behavior and select the pretraining data using that cheaper model. We show the effectiveness of this approach in two dataset selection settings: language modeling and imitation learning. Second, we explore the role of data in model reliability and consider two different threat models: backdoor attacks and malicious data editing. In this first threat model, an adversary injects a few doctered samples into the training set to control model predictions at inference time. We study the effect of these malicious samples on model behavior and then propose a framework for detecting and removing them from the training data. In the second threat model, an adversary leverages generative models such as diffusion models to maliciously modify personal data and generate harmful digital content. We focus on image editing and investigate how we can imperceptibly modify personal images to mitigate editing using diffusion models and raise and the cost of hamrful content generation. Overall, this thesis contributes to the understanding of the role of the data in driving model behavior. Through these efforts, we aim to provide mechanisms for (i) training models that perform better and (ii) are more reliable when deployed in the real world.

Understanding and Overcoming Optimization Barriers in Non-convex and Non-smooth Machine Learning

Mon, 01 Sep 2025 00:00:00 GMT

Understanding and Overcoming Optimization Barriers in Non-convex and Non-smooth Machine Learning Gatmiry, Khashayar At their core, our machine learning systems are trained by solving an optimization problem, where the goal is to minimize a predefined objective function by adjusting model parameters based on the data. Despite the wealth of structure and prior knowledge present in the data and feedback, our training methods remain relatively simple and independent of this structure. In spite of, or perhaps because of, this simplicity, these methods are often lacking in theoretical guarantees. To design machine learning algorithms that are less data-hungry while ensuring theoretical guarantees on both computational efficiency and output validity, it is essential to better understand and leverage the rich structure within the learning setup and the data distribution, e.g. by altering the geometry of the solution space or adjusting the objective function to induce a more effective learning procedure. This approach moves beyond classical algorithm design, which focuses primarily on handling worst-case instances. This thesis investigates the optimization landscape of central learning problems and develops geometric and analytic schemes adapted to their structure, leading to algorithms with superior computational and statistical performance. In addition, it seeks to advance our mathematical understanding of the principles underlying the success of deep learning.

Seeing the Forest Through the Trees: Knowledge Retrieval for Streamlining Particle Physics Analysis

Mon, 01 Sep 2025 00:00:00 GMT

Seeing the Forest Through the Trees: Knowledge Retrieval for Streamlining Particle Physics Analysis McGreivy, James C. Generative Large Language Models (LLMs) are a promising approach to structuring knowledge contained within otherwise unmanageable corpora of research literature produced by large-scale and long-running scientific collaborations. Within experimental particle physics, such structured knowledge bases could expedite methodological and editorial review. Complementarily, within the broader scientific community, generative LLM systems grounded in published work could make for reliable companions allowing non-experts to analyze openaccess data. Techniques such as Retrieval Augmented Generation (RAG) rely on semantically matching localized text chunks, but struggle to maintain coherent context when relevant information spans multiple segments, leading to a fragmented representation devoid of global cross-document information. In this work I utilize the hierarchical organization of experimental physics articles to build a tree representation of the corpus, and present the SciTreeRAG system which leverages this structure with the aim of constructing contexts more focused and contextually rich than a standard RAG. Additionally, I develop methods for using LLMs to transform the unstructured corpus into a structured knowledge graph representation. I then implement SciGraphRAG, a retrieval system that leverages this knowledge graph to access global cross-document relationships eluding standard RAG, with the goal of encapsulating domain-specific connections and expertise. I demonstrate proof-of-concept implementations of both systems using the corpus of the LHCb experiment at CERN.

Development of Low-Cost In Situ Gas Sensors for Oceanographic Applications

Mon, 01 Sep 2025 00:00:00 GMT

Development of Low-Cost In Situ Gas Sensors for Oceanographic Applications Gower, Elizabeth Ann Anthropogenic activity has increased atmospheric carbon dioxide (CO₂) levels, disrupting the global carbon cycle and driving widespread environmental change. The ocean acts as a major sink. Accurate and scalable in situ monitoring of oceanic carbon chemistry is vital for understanding the impacts of climate change and informing marine carbon dioxide removal (mCDR) strategies. Many existing in situ instruments for marine applications are constrained by their size, cost, power requirements, or reliance on consumable reagents. Developing low-cost, compact, low-power, and accurate in situ sensors would significantly enhance the spatiotemporal resolution of oceanographic data and enable widespread monitoring of dissolved gases throughout the ocean. This, in turn, would deepen our understanding of how, where, and when changes are occurring within the marine carbon cycle. Two key variables essential for studying this cycle are the partial pressure of carbon dioxide (pCO₂) and dissolved inorganic carbon (DIC). This thesis presents the development of two sensors, one for in situ pCO₂ measurement and another for novel DIC quantification, both designed to be affordable, reliable, and scalable tools for advancing our understanding of ocean chemistry and the global carbon system. First, the development, calibration, and open-ocean deployment of a miniaturized Dissolved Multi-Gas Sensor (DMGS) that measures pCO₂ and partial pressure of oxygen (pO₂) is presented. The sensor was integrated into a custom-built surface drifter designed to entangle with Sargassum mats and send data autonomously. The drifter utilized commercial off-theshelf (COTS) components and cost roughly $1000 to build. After lab testing, a drifter was deployed in the Great Atlantic Sargassum Belt (GASB) and collected data for 22-days. In addition to gas data, the drifter tracked temperature, light intensity, humidity, pressure, and location sending measurements via an Iridium satellite. The resulting data captured dynamic changes in localized gas concentrations, temperature, and light levels that highlighted photosynthetic and respiratory activity within Sargassum patches. These drifters demonstrate the value of in situ data to investigate marine biogeochemical processes that contribute to the marine carbon cycle, especially in areas with high biologic activity. Next, this thesis presents the iterative development of a novel DIC sensor with potential for future in situ applications. Initial prototypes tested the feasibility of using a COTS CO2 sensor in both static and flow-through configurations, however sensor saturation issues prompted a shift to a pressure-based detection method. Multiple test setups were evaluated for pressure stability and sensor sensitivity, culminating in a bottle-based flow system that demonstrated the potential for reagent-minimized, pressure-based DIC quantification. With the final setup, a COTS pressure sensor that sat behind a gas permeable membrane was found to repeatably and accurately quantify DIC from acidified seawater. This approach of quantifying DIC via pressure change is novel in the field of gas sensing and maintains a low-cost, accessible design. Together, the sensors developed in this thesis expand the toolkit for marine carbon monitoring and provide a foundation for affordable, distributed sensing networks. These technologies enable higher-resolution insights into ocean biogeochemistry and support critical monitoring, reporting, and verification (MRV) frameworks needed to evaluate the effectiveness of mCDR techniques. Continued refinement of these low-cost platforms could play a key role in understanding and mitigating anthropogenic impacts on marine systems.

Applying Reference Class Forecasting to Multifamily Investments: Identifying and Capturing Operational Alpha through the Outside View

Mon, 01 Sep 2025 00:00:00 GMT

Applying Reference Class Forecasting to Multifamily Investments: Identifying and Capturing Operational Alpha through the Outside View Firouzian, Fardean This thesis applies Reference Class Forecasting (RCF) to multifamily real estate underwriting as a means of countering optimism bias, strategic misrepresentation, and other distortions embedded in the traditional “inside view.” Adapted from its proven application in infrastructure and corporate capital budgeting, RCF anchors projections in the actual performance distributions of comparable assets rather than in deal-specific narratives. The research centers on the development of the “Comp Warehouse,” a structured repository of property-level financials organized by market, asset class, vintage, and unit scale. By benchmarking assumptions against statistically valid reference classes, the approach enforces empirical discipline and highlights opportunities for “operational alpha”—the marginal increase in net operating income (NOI) achieved when underperforming assets converge on median peer performance. A South Florida case study demonstrates the method’s utility in an acquisition context. Analysis of 48 assets across Melbourne, Miami, Fort Lauderdale, and West Palm Beach shows that while rent levels cluster tightly around market medians, operating expenses vary widely, producing large dispersion in realized NOI. Applying the framework to a 191-unit Class A property in Fort Lauderdale illustrates how RCF can ground underwriting assumptions by distinguishing between defensible revenue-driven growth strategies and less plausible expense-reduction projections proposed in a bidding scenario. Recognizing constraints of both scale and frequency, this thesis also explores artificial intelligence as a tool for automating the ingestion and standardization of operating statements and rent rolls. Properly deployed in a human-in-the-loop framework, AI can reduce data friction, expand sample sizes, and sharpen forecasting precision. The contribution of this thesis is twofold: it demonstrates the feasibility of applying RCF to the multifamily sector—an asset class whose relative standardization, liquidity, and data availability make it especially conducive to outside-view benchmarking—and it situates the methodology within a technology-native architecture designed to scale empirical discipline, enhance underwriting rigor, and systematically capture operational alpha.

Concretely-Efficient Multi-Key Homomorphic Secret Sharing and Applications

Mon, 01 Sep 2025 00:00:00 GMT

Concretely-Efficient Multi-Key Homomorphic Secret Sharing and Applications He, Kaiwen Homomorphic secret sharing (HSS) is a powerful cryptographic primitive that enables efficient, low-communication secure computation without the use of fully homomorphic encryption. Public-key HSS is a well-known variant that supports inputs from multiple parties, but all parties must agree on a joint public key before any party can encode their inputs, requiring extra rounds of communication in applications. Recently, Couteau et al. (EUROCRYPT 2025) constructed multi-key HSS (MKHSS)—a new primitive which allows parties to encode their inputs under independent keys—under the DCR assumption. MKHSS assumes only a reusable common reference string, without the need for prior interactions between parties or a public-key infrastructure. In this paper, we construct and implement the first concretely-efficient MKHSS scheme under the same assumptions used by Couteau et al. Using an algorithmic insight that reduces the largest modulus in Couteau et al. from N⁴ to N², our optimized implementation can homomorphically multiply inputs in 5.0 milliseconds—while an implementation of Couteau et al. requires 224.6 milliseconds—thereby achieving a 45× speedup. A powerful application of MKHSS is to realize attribute-based non-interactive key exchange (ANIKE), which generalizes password-based key exchange (PAKE) to arbitrary attribute policies. ANIKE is currently only known from MKHSS. We use our implementation to evaluate the first concretely-efficient ANIKE schemes for a range of practically useful policies. Using our implementation, two parties can perform a geolocation-based key exchange in 1.65 seconds and a fuzzy PAKE on an 8-word passphrase in 7.59 seconds for realistic parameters, on a single core. Compared to using Couteau et al., which requires 62.5 and 253 seconds, we achieve 38× and 33× speedups, respectively.

Reciprocity and Normality in the Scattering Matrix of Disordered Media

Mon, 01 Sep 2025 00:00:00 GMT

Reciprocity and Normality in the Scattering Matrix of Disordered Media Bharadwaj, Shreyas K. The scattering matrix formalism provides a practical characterization of wave transport in linear, source-free systems by relating a set of operationally defined input and output spatial channels. The matrix is structured as a block operator, with diagonal blocks encoding same-side reflection matrices (RMs) and off-diagonal blocks encoding transmission matrices (TMs) in opposing propagation directions. Under Helmholtz reciprocity, symmetry relations are imposed: RMs are symmetric, and forward and reverse TMs are mathematical transposes of each other. These relations were employed as constraints to correct system-induced aberrations in measured scattering matrices of complex optical media via a matrix-based gradient descent procedure. Resulting phase corrections corresponded closely with classical aberration modes without heuristic parameterizations, suggesting that these modes naturally arise to restore reciprocity-induced symmetry. Vectorial TMs were measured for single- and double-pass propagation through step-index MMFs and scattering samples, with corrected phase terms showing agreement across sample types. Furthermore, matrix normality was introduced as a descriptor of stable modal transport. Normal matrices admit unitary diagonalization, reflecting orthogonal eigenchannels and spectrally coherent propagation. Near-normal behavior was observed in fiber TMs, while RMs of scattering slabs remained strongly non-normal, as quantified by a normalized Henrici departure. Sufficient conditions for normality were identified in terms of the system Green’s function and its bi-compression onto the measurement basis. A complementary dispersion experiment investigated two regimes: nearly-normal MMFs, where the Wigner–Smith time-delay operator was jointly diagonalizable and supported accurate first-order spectral models; and mechanically compressed fibers, where loss of normality produced noncommuting operators and collapse of model fidelity. These results suggest that normality captures well-behaved modal transport, underpinning the validity of parametric models and other operator-based analyses of disordered media. Together, reciprocity and normality impose complementary constraints on wave transport: reciprocity governs global symmetry, while normality captures internal coherence of modal propagation. Relevance is noted for matrix-based imaging, inverse scattering theory, and non-Hermitian wave physics, where symmetry and modal stability remain central.

Mesh Differentiable Rendering for Real-World Scenes

Mon, 01 Sep 2025 00:00:00 GMT

Mesh Differentiable Rendering for Real-World Scenes Charatan, David Differentiable rendering has established itself as an effective tool for 3D reconstruction and novel view synthesis. Most state-of-the-art differentiable rendering methods use purpose-built renderers to optimize specialized, nonstandard 3D representations. However, most downstream applications of differentiable rendering rely on 3D meshes, which are near-universally supported due to their suitability for a wide range of rendering, simulation, and 3D modeling workflows. While prior methods have explored using 3D meshes directly within gradient-based optimization, they have been limited to object-centric scenes and cannot reconstruct real-world, unbounded scenes. This work addresses this shortcoming via a differentiable rendering formulation that combines an off-the-shelf, non-differentiable triangle rasterizer with a 3D representation that consists of nested mesh shells. During every forward pass, these shells are extracted from an underlying signed distance field. Then, the shells are independently rasterized and the resulting images are alpha-composited using opacities derived from the shells' per-vertex signed distance values. Notably, the shells' vertex positions are updated only via the underlying signed distance field, not via backpropagation through the rasterizer itself. This makes our method compatible with off-the-shelf, non-differentiable triangle rasterizers. To the best of our knowledge, our method is the first differentiable mesh rendering method that scales to unbounded, real-world 3D scenes, where it produces high-quality novel view synthesis results whose quality approaches the quality of state-of-the-art, non-mesh-based methods. Our method's performance is also competitive with state-of-the-art surface rendering methods on object-centric scenes. Ultimately, our method suggests that it may be possible to solve the differentiable rendering problem using tools from the conventional graphics toolbox rather than relying on specialized renderers.

Task-Aware Spatial and Temporal Aggregation for Capacity Expansion Planning

Mon, 01 Sep 2025 00:00:00 GMT

Task-Aware Spatial and Temporal Aggregation for Capacity Expansion Planning Duguey, Gabriel As we plan tomorrow’s electricity system, we face fundamental questions: where should new power plants go, which technologies deserve investment, and how much transmission is enough? These decisions are the domain of Capacity Expansion Planning (CEP), a class of optimization models that guide long-term infrastructure investments in power systems. To be realistic, CEP models must capture fine-grained spatial and temporal variations because demand varies by city and climate, while wind and solar output depend on weather patterns that shift hour by hour and location by location. But representing the system with thousands of time steps and hundreds of nodes makes the optimization problem computationally too large to solve. This thesis addresses the core question: how can spatial and temporal aggregation in CEP models be designed to preserve planning-relevant patterns that drive investment decisions? Existing approaches often treat aggregation as a neutral preprocessing step, relying on heuristics like political boundaries or geographic proximity. In contrast, we propose a task-aware pipeline that treats aggregation as an integral modeling decision, explicitly aligned with planning objectives. The approach builds a composite similarity metric that blends diverse planning-relevant signals, including, but not limited to, duration curves, ramping behavior, and spatial correlation, and uses k-medoids clustering to define spatial zones. Temporal aggregation is then applied to daily system-wide profiles, selecting representative days that maintain cross-zonal interactions. The result is a reduced spatio-temporal dataset fed into a CEP model. The resulting investment decisions are re-evaluated at full resolution to evaluate their feasibility and real cost. Experiments on a New England case study show the pipeline consistently outperforms common baselines like political boundaries, geographic proximity, or capacity factor statistics. Among 50 feature weightings, the best design reduces system cost by 13% compared to heuristics. Correlation-based features drive the best results, while raw amplitude and geographic location often degrade performance when used alone.

Earth as Equity Partner: A Revolutionary Approach to Ecological Conservation through Housing Development

Mon, 01 Sep 2025 00:00:00 GMT

Earth as Equity Partner: A Revolutionary Approach to Ecological Conservation through Housing Development McDonough, Kate Duddington Farm is a 312-acre site north of Baltimore, Maryland. A stream restoration project was completed at the location nearly a decade ago in concert with the State of Maryland, the Manor Conservancy, Ecotone, and landowners Harry and Tara McDonough. The project was conducted with some success, however due to a lack of State oversight and long-term management provisions, the ecology has since declined. The following proposal outlines a new model for long-term land restoration and conservation, whereby land conservation and restoration are financed not solely through short term grants and fragile easements, but through the thoughtful use of modest real estate interventions. A small cluster of homes is developed on one portion of the site. The act increases the value of the land, generates equity, and establishes a permanent conservation fund. The design protects habitat and invites people into a deeper relationship with the natural world. The plan offers scalability in taking the land value capture and applying it to future land conservation projects, compounding returns and projecting a model to preserve hundreds of thousands of acres of critical land across the United States. This model highlights Indigenous ecological knowledge (TEK) and traditional practices of engaging with the land, highlighting a deeper understanding of how humans and nature can coexist in mutually healthy ways. The model is designed at a time when watersheds, national parks, and old-growth forests are faced with the greatest threat to global ecology. Duddington Farm is used as a retrospective case, but the broader goal is to create a regenerative framework for conservation-based development across critical watershed regions.

The Causal Effects of Mandatory Quarterly Earnings Guidance on Corporate Information Environment and Corporate Short-Termism

Mon, 01 Sep 2025 00:00:00 GMT

The Causal Effects of Mandatory Quarterly Earnings Guidance on Corporate Information Environment and Corporate Short-Termism Wang, Yuting I examine the causal effects of mandatory quarterly earnings guidance using a regulatory mandate in China that required a subset of listed firms to issue bundled quarterly earnings guidance from 2007 to 2018. A difference-in-differences analysis shows that when these firms are no longer required to issue such guidance, their corporate information environment deteriorates, evidenced by reduced analyst coverage, fewer site visits, and lower price timeliness, meaning that stock prices incorporate less information about current and future earnings. However, these firms increase R&D and SG&A spending, consistent with alleviated managerial myopia as short-term market pressure eases. These findings highlight the dual-edged nature of the mandatory quarterly earnings guidance and offer insights for both practitioners and policymakers.

Automated Finite Elements

Mon, 01 Sep 2025 00:00:00 GMT

Automated Finite Elements Collin, Teodoro Fields Finite element methods (FEMs) are a powerful and ubiquitous tool for solving engineering problems. Experimenting with different finite elements can improve the quality and efficiency of solutions. Furthermore, in some cases, the wrong (but nonetheless most common) choice of finite element will produce solutions which converge to the wrong answer regardless of mesh resolution. However, in practice, the choice of finite element is not explored due to the complexity of re-deriving and re-implementing finite element methods. Trying a new finite element is challenging because practitioners must manually deduce formulas to use these elements and they must implement these formulas within the context of a potentially complex system. We address this problem by introducing ElementForge, a finite element system that is parametric over the literate mathematical specification of a finite element in a domain-specific language (DSL). The ElementForge compiler reasons about tensor spaces, tensors, and tensor bases from first principles to derive implementations of finite elements. The ElementForge compiler is able to automatically derive implementations of finite elements previously only derived by hand. Further, ElementForge minimally couples several key mathematical concepts, mainly tensor fields, mesh topologies, sparse tensors, and assembled finite element operators, to produce a complete finite element system that is parametric over the choice of element. Consequently, the elements derived by the compiler can be applied parametrically to new meshes, PDEs, and boundary conditions. We evaluate our system by implementing several simulations with different finite elements, demonstrating that our system can explore tradeoffs in generality, accuracy, speed, and representational complexity. For example, we are able to implement the Morley, Bell, Argyris, and Hermite like elements with less than 50 lines of code and use them all in a single simulation.

Bespoke Threat Models: Achieving Realistic Privacy Guarantees for Deployed Protocols

Mon, 01 Sep 2025 00:00:00 GMT

Bespoke Threat Models: Achieving Realistic Privacy Guarantees for Deployed Protocols Hogan, Kyle This thesis focuses on the question of what degree of privacy is achievable in the real world for long-running applications. We explore this question in two main settings: private advertising and anonymous communication. In doing so, we consider constraints each application may have in practice and what adversarial model is realistic for the context in which the application will be deployed. For real world applications, achieving perfect privacy — especially against a worst case adversary — can be impossible. That is, perfect privacy, while achievable in theory, may in practice require assumptions that conflict with usability, deployability, or utility requirements. This presents a challenge as privacy-preserving technologies can, necessarily, only provide privacy for the people who use them. Because of this, designing around user experience is critical, even if doing so requires compromises in the theoretical degree of privacy a system can provide or the strength of adversaries considered in its threat model. In the space of private advertising, we first propose a novel protocol, AdVeil, that eliminates leakage of user data beyond that revealed by the input/output of the ads ecosystem as a whole. We then provide a minimal modeling of the functionality of digital advertising which we use to prove that, even for systems like AdVeil with minimal leakage, the advertising metrics released at the end of the protocol are sufficient to leak information about end users to advertisers when combined with their audience targeting criteria. In the space of anonymous communication, we propose ShorTor, a new routing protocol for Tor that utilizes techniques popular with content distribution networks (CDNs) to reduce latency while maintaining Tor’s existing anonymity guarantees. We evaluate this protocol using a dataset of over 400,000 latency measurements we collected between the 1,000 most popular Tor relays.

Operationalizing Reliable Machine Learning: From Data Collection to Model Presentation

Mon, 01 Sep 2025 00:00:00 GMT

Operationalizing Reliable Machine Learning: From Data Collection to Model Presentation Balagopalan, Aparna Automated systems driven by machine learning (ML) have made exciting progress across a spectrum of applications. Despite such progress, encoded biases and other failure modes may create barriers to the real-world utility and reliability of such systems. For example, nonrandom data missingness, biased algorithmic optimization objectives, or model presentation strategies that incorrectly impact user trust can all cause models to fail in practice. In this thesis, guided by such observations and prior work on pipeline-awareness in machine learning, we aim to operationalize reliable ML. Under this goal, we propose a framework consisting of the following three components: responsible data collection, robust algorithm development, and fair model presentation. We first conduct two case studies to advance responsible data collection. We investigate whether standard procedures for acquiring data can be repurposed when training models to mimic human judgments about norm violations. We also demonstrate patterns of delayed demographic data reporting within a longitudinal healthcare dataset and show that timevarying missingness due to such delays can distort disparity assessments. Second, we introduce two novel algorithms to improve reliability: a method that leverages representations from vision-language models to filter noisy training data, and a method to produce fair rankings that account for properties of search queries. Finally, since the presentation design of predictions impacts trust in model consumers, we propose metrics to quantify the fairness of post-hoc explainability techniques. Thus, with this thesis, we re-evaluate measurements throughout the machine learning pipeline and contribute to the broader goal of reliable machine learning.

Anti-phage defense as a driver of molecular innovation

Mon, 01 Sep 2025 00:00:00 GMT

Anti-phage defense as a driver of molecular innovation Doering, Christopher Ross Bacteriophages, or phages for short, pose a near-constant threat to the bacteria they infect. Billions of years of conflict has been a catalyzing force for the creation of bacterial defense systems and corresponding phage evasion strategies. To counter phage predation, bacteria have developed a vast diversity of enzyme chemistries and molecular sensing mechanisms whose study has produced new biotechnological tools and insights into our own immune systems. In this work, I have investigated anti-phage defense mechanisms at multiple scales using a combination of genetic, biochemical, and bioinformatic approaches. First, I characterized the mechanism of action of the anti-phage defense system CmdTAC, a toxin-antitoxin-chaperone system that recognizes a viral structural protein to activate a novel mRNA ADP-ribosyltransferase, thereby halting infection. Next, I examined the diversity and distribution of anti-phage mechanisms encoded by E. coli lysogenic phages – phages capable of integrating into and lying dormant within their bacterial hosts. This analysis uncovered overlooked classes of lysogenic phages harboring novel candidate defense systems, including one newly validated system with no detectable homology to previously known mechanisms. Together, this work broadens our understanding of bacterial immune systems, expands the pool of known enzyme chemistries, and highlights areas where continued study can reveal additional mechanisms of anti-phage defense.

Shaping Function Through Space: The Role of Spatial Organization in Microbial Communities

Mon, 01 Sep 2025 00:00:00 GMT

Shaping Function Through Space: The Role of Spatial Organization in Microbial Communities Toneatti Vercelli, Gabriel Spatial organization plays a critical role in microbial community function, influencing how cells exchange metabolites, coordinate behavior, and compete for resources. This thesis investigates the consequences of spatial structure in natural microbial systems and introduces a novel method to engineer these systems with high precision and scalability. First, we examine the colonization of chitin particles by marine bacteria, a model for particulate organic matter degradation. Using high-throughput phenotyping of natural isolates, we show that vitamin cross-feeding is essential for successful colonization of chitin-particles by many auxotrophic strains. We then model two distinct vitamin cross-feeding mechanisms: lysis and secretion. Using a resource-explicit modeling approach, we leverage metabolic-flux and physiological measurements to predict the colonization success of auxotrophic cross-feeders in this spatially structured environment. Second, we introduce a new chemical method for engineering microbial cell surfaces that enables covalent attachment of molecules such as enzymes and DNA strands to the cell surface. We show that this surface functionalization procedure leads to the acquisition of new phenotypes like antibiotic resistance and programmable adhesion. Altogether, this work reinforces the importance of spatial organization for microbial community function and introduces a new technique to harness this community feature and turn it into a design principle for synthetic microbial systems.

LumiModeling: A Gaussian Splatting-Based Tool for Recreating Dynamic Material and Lighting Interaction in Architecture

Mon, 01 Sep 2025 00:00:00 GMT

LumiModeling: A Gaussian Splatting-Based Tool for Recreating Dynamic Material and Lighting Interaction in Architecture Cao, Biru This thesis presents LumiModeling, a real-time visualization tool based on Gaussian Splatting (GS) that simulates the dynamic interplay between materiality and lighting in architectural environments. While conventional design workflows rely on geometric modeling and photorealistic rendering, they often abstract complex material behaviors and fall short in capturing light-material interactions. In contrast, GS enables the reconstruction of high-fidelity 3D models from 2D image sets, representing viewdependent effects such as reflection, transparency, and surface roughness. A comparative analysis using real-world data from the MIT Stata Center and the Met Warehouse demonstrates GS’s advantages over mesh-based photogrammetry, particularly in rendering reflective and transparent materials. This work extends existing GS capabilities by implementing a relightable pipeline based on the existing model Relightable3DGaussian (Gao et al., 2023), in which each Gaussian point is augmented with physical parameters, including BRDF, surface normals, and incident lighting. The Stata Center dataset is used to test the relighting of GS. A user study involving architecture professionals reveals that perceptual focus shifts from geometry to materiality and lighting as visual realism increases. The findings highlight the potential of relightable GS in architectural visualization and anticipate its integration into future design workflows.

Urban Data Memory: Using Generative AI to Structure and Visualize Zoning Data for Urban Planning Evaluation

Mon, 01 Sep 2025 00:00:00 GMT

Urban Data Memory: Using Generative AI to Structure and Visualize Zoning Data for Urban Planning Evaluation Kupershmidt, Adi Urban planners face significant challenges in systematically and quantitatively evaluating past planning practices, stemming, among other reasons, from the scarcity of accessible structured data. The period from a plan’s initiation to implementation can span generations; recorded data from the planning processes are often deemed obsolete for addressing present concerns by the time of post-occupancy evaluation. This research examines whether generative AI can help bridge this gap and under what conditions - highlighting both challenges and opportunities - by introducing a system that responsively transforms qualitative zoning data into structured, queryable formats to support the quantitative analysis of planning practices. A database of ~150 approved semi-structured urban plans under Tel Aviv municipality’s local jurisdiction supports this project's case study. The system relies on proprietary LLMs (ChatGPT, Claude), streamlining a natural language query input through 3 agentic tasks: (1) RAG (Retrieval Augmented Generation) based querying, generating free-text answers from all plans, (2) structuring the answers to a valid JSON, and (3) visualizing structured data. Key findings indicate an 85.45% precision of the system, as evaluated through an end-to-end assessment of 11 representative queries, each validated against 40 manually labeled plans. The tool provides actionable insights, enabling queries such as trends in sheltered bicycle parking approvals or the status of affordable housing planning over the past decade. This research underlines the significance of flexibly structuring non- and semi-structured data for urban science. It addresses the growing gap between static legacy data collection and real-time policymaking, democratizing access to planning information and fostering informed decision-making practices. Integrating cutting-edge AI-driven tools contributes to the current discourse on AI applications for city management and planning by providing a replicable model for more cities and planning datasets to build upon and improve.

Developing Methods for Enhanced Measurement of DNA Single-Strand Breaks and Somatic Variants

Thu, 01 Feb 2024 00:00:00 GMT

Developing Methods for Enhanced Measurement of DNA Single-Strand Breaks and Somatic Variants Elacqua, Juniper J. Maintenance and repair of DNA are essential for proper cellular functioning and preventing the emergence of disease states. As cells divide, mutations accumulate in the genome which contributes to aging phenotypes and can result in genetic diseases such as cancer. The rate at which a cell develops mutations can be accelerated through exposure to genotoxic agents that introduce lesions which, if left unrepaired, prevent accurate replication of the genome. As such, it is crucial to understand the ways in which DNA becomes damaged, how cells respond to various types of damage, and how this damage contributes to mutagenesis and the development of genetic disease. These fields of study have been greatly advanced by improvements in DNA sequencing technologies, and here we present two sequencing-based methods that aim to enable deeper study of DNA damage, repair, and mutagenesis. First, we demonstrate DENT-seq, a method that identifies single-strand breaks with single-nucleotide resolution. Single-strand breaks are the most common form of DNA damage, occurring at rates of ~10,000 per cell per day, but have to date been understudied due to lack of an unbiased, high-resolution method for their detection. Second, we improve upon lineage sequencing, a previously reported method that uniquely measures somatic single nucleotide variants in dividing cells to achieve high specificity/sensitivity as well as the ability to temporally resolve variants and to relate sequenced genotypes to optically observed cellular phenotypes. Despite the high-quality data and unique capabilities offered by this method, it has so far been underused due to a need for complex, microfluidic-based cell collection. We demonstrate novel protocols for performing lineage sequencing that enable easy adoption of the method without the need for highly specialized equipment or expertise. In addition, we expand the repertoire of mutations measurable with the technique to include indels and variants that arise specifically in response to a genotoxic treatment. The methods we show can be applied to reveal novel findings regarding the causes and consequences of DNA damage and mutagenesis that underly numerous genetic diseases.Maintenance and repair of DNA are essential for proper cellular functioning and preventing the emergence of disease states. As cells divide, mutations accumulate in the genome which contributes to aging phenotypes and can result in genetic diseases such as cancer. The rate at which a cell develops mutations can be accelerated through exposure to genotoxic agents that introduce lesions which, if left unrepaired, prevent accurate replication of the genome. As such, it is crucial to understand the ways in which DNA becomes damaged, how cells respond to various types of damage, and how this damage contributes to mutagenesis and the development of genetic disease. These fields of study have been greatly advanced by improvements in DNA sequencing technologies, and here we present two sequencing-based methods that aim to enable deeper study of DNA damage, repair, and mutagenesis. First, we demonstrate DENT-seq, a method that identifies single-strand breaks with single-nucleotide resolution. Single-strand breaks are the most common form of DNA damage, occurring at rates of ~10,000 per cell per day, but have to date been understudied due to lack of an unbiased, high-resolution method for their detection. Second, we improve upon lineage sequencing, a previously reported method that uniquely measures somatic single nucleotide variants in dividing cells to achieve high specificity/sensitivity as well as the ability to temporally resolve variants and to relate sequenced genotypes to optically observed cellular phenotypes. Despite the high-quality data and unique capabilities offered by this method, it has so far been underused due to a need for complex, microfluidic-based cell collection. We demonstrate novel protocols for performing lineage sequencing that enable easy adoption of the method without the need for highly specialized equipment or expertise. In addition, we expand the repertoire of mutations measurable with the technique to include indels and variants that arise specifically in response to a genotoxic treatment. The methods we show can be applied to reveal novel findings regarding the causes and consequences of DNA damage and mutagenesis that underly numerous genetic diseases.

Reconfigurable and Interference-Tolerant Receivers for Next Generation Wireless Systems

Mon, 01 Sep 2025 00:00:00 GMT

Reconfigurable and Interference-Tolerant Receivers for Next Generation Wireless Systems Araei, Soroush An “all-in-one” radio, programmable across the sub-7 GHz spectrum, offers significant hardware efficiency for 5G systems. However, addressing strong interferers in this wide and congested spectrum remains a major design challenge. N-path filters offer a promising solution for efficiently suppressing interference, thanks to their clock-controlled reconfigurability and excellent linearity against in-band and adjacent-channel blockers. While widely adopted in modern receiver architectures, these switched-capacitor circuits remain inherently vulnerable to blockers at clock harmonics, due to their hard-switching nature. These blockers, common in 5G bands, pose a key bottleneck, delaying the realization of fully integrated multi-band, multi-mode radios. This dissertation introduces fully passive topologies to address this challenge. The first design leverages simultaneous charge sharing and capacitor stacking to implement harmonic rejection filtering. It operates entirely without active circuitry and exhibits exceptionally low loss. A second-generation technique, termed “harmonic reset switching”, builds on this approach by rejecting harmonic blockers directly at the driving point of the N-path filter, achieving superior performance with reduced circuit complexity. As a result, existing reconfigurable receiver topologies can be seamlessly transformed into harmonic blocker–resilient architectures. For example, a taped-out mixer-first receiver adopting this technique achieves a 100× improvement in third-harmonic blocker tolerance compared to state-of-the-art broadband receivers. This dissertation also proposes a reconfigurable receiver for IoT-class radios that is tolerant to both close-in and far-out blockers. A scalable clock bootstrapping technique is introduced to enhance linearity while maintaining both power and cost efficiency. All designs are validated through prototypes fabricated in advanced 22-nm and 45-nm silicon-on-insulator (SOI) technologies. By addressing this long-standing challenge, this work paves the way for fully reconfigurable, interference-resilient radios for 5G and beyond.

Video as the Language of Embodied Intelligence

Mon, 01 Sep 2025 00:00:00 GMT

Video as the Language of Embodied Intelligence Chen, Boyuan Achieving general-purpose embodied intelligence remains a central challenge in artificial intelligence. While recent efforts have extended Large Language Models (LLMs) to robotics by incorporating additional modalities, these adaptations face critical limitations in perception, grounding, and control. For example, spatial reasoning—a simple yet indispensable capability for robots—reveals one of such shortcoming clearly: multimodal LLMs often fail even basic spatial perception tasks like estimating distances. This thesis begins by examining these failures through SpatialVLM, a system that augments vision-language models with 3D spatial reasoning. Although more effective in spatial estimation, this work reveals a deeper issue: the fundamental expressive limitations of language-only outputs in capturing sensorimotor dynamics. Based on these findings, the thesis advocates for a ground-up methodology for robot foundation models, starting with identifying an appropriate “language” for embodied AI, then architecting models and training regimes accordingly. We investigate video as the foundational language, integrated with model-based planning for decision-making. This new paradigm is instantiated through two core contributions. The first is Diffusion Forcing, a hybrid modeling framework that combines causal next-token prediction with full-sequence diffusion. This approach supports stable, coherent rollouts far beyond the training horizon and allows guided generation for decision-making tasks, bridging predictive modeling and planning. Building on Diffusion Forcing, we introduce the Diffusion Forcing Transformer (DFoT), a natural architectural extension designed for flexible video generation conditioned on variable-length histories. To further support long-horizon world-modeling, we propose History Guidance, a set of techniques that enhance sample fidelity, temporal consistency, and compositional generalization. Together, these methods enable robust modeling of visual dynamics across extended timeframes. Finally, we present a preliminary yet promising video foundation model for zero-shot robot motion planning, highlighting the potential of video as the foundational language of embodied intelligence.

Biologically Interpretable Representation Learning for Mechanistic Insights into Cancer Immunotherapy Resistance

Mon, 01 Sep 2025 00:00:00 GMT

Biologically Interpretable Representation Learning for Mechanistic Insights into Cancer Immunotherapy Resistance Tariq, Ifrah Resistance to immune checkpoint inhibitors (ICIs) remains a critical barrier to effective cancer therapy, driven by complex, multi-scale interactions that current biomarkers often fail to capture. This dissertation introduces the Biologically Disentangled Variational Autoencoder (BDVAE)—an interpretable deep learning framework designed to uncover mechanistic drivers of ICI resistance through multi-omic data integration. Using RNA-seq and wholeexome sequencing data from 366 patients across melanoma, renal cell, urothelial, and gastric cancers, BDVAE learns low-dimensional latent representations that are both predictive of response and biologically meaningful. The model reveals distinct latent dimensions aligned with immune regulation, tumorintrinsic signaling, metabolism, and neuroimmune interactions. SHAP-based interpretation and pathway analysis highlight key resistance-associated programs, including immunosuppressive cytokine signaling, metabolic signaling, and neuroactive pathways such as calcium and cAMP signaling. Unsupervised clustering identifies three tumor subtypes—responder-dominant, non-responder-dominant, and an intermediate group—suggesting plastic or transitional immune states. Survival analyses confirm the clinical relevance of these clusters and expose heterogeneity within standard RECIST categories. Overall, this work presents a novel, interpretable framework for modeling ICI response, offering insights into resistance mechanisms and actionable paths for biomarker discovery, patient stratification, and therapeutic innovation in precision immuno-oncology.

Geometric interpretations of structural demand for the analysis and reduction of design complexity

Mon, 01 Sep 2025 00:00:00 GMT

Geometric interpretations of structural demand for the analysis and reduction of design complexity Lee, Keith Janghyun This dissertation presents a computational framework to effectively interpret the distribution of structural demand that emerges from the design of large-scale structural systems, and develops methods for its quantification and manipulation. Structural demand is the required strength and geometry of individual building components that emerges from design as a result of global geometry, topology, and loading. Existing metrics of structural performance fail to consider how variations in demand at the component level can lead to designs that are theoretically efficient but difficult to construct. This has led to a rejection of low-carbon, high-performance design solutions in practice, or the need for extensive post-hoc rationalization, both under the presumption of untenable design complexity for conventional building practices. This dissertation argues that an explicit consideration of the distribution of induced structural demand can bridge this gap between design intent and construction feasibility. To achieve this, structural demand is interpreted as sets of geometric objects in n-dimensional feature spaces, where each dimension represents an independent component of demand, such as area, length, or stiffness. By directly visualizing the spatial distribution of demand, designers are presented with a richer context of non-physical structural design information, and can evaluate how decisions in structural form affect this distribution. Further, spatial interpretations of information allow for spatial metrics of similarity and variation to be defined, from which quantitative measures of design complexity are derived that account for the shape and distribution of demand. This framework, named \emph{Demand Space Analysis}, is explored in depth and applied to a range of structural scales, from the demand of truss elements and their connections, to the relationship between demand and fixed sets of capacity. Advancements in structural optimization are also presented, enabling more efficient and direct minimization of modern structural performance metrics, from which the relationship between design performance and demand complexity can be explored. Through case studies in each chapter, this dissertation demonstrates how geometric analysis of structural demand information can inform the designer of the implications of decisions on the perceived complexity of design, and provides tools for its quantification and reduction.

Behavioral Responses to Congestion Pricing in New York City: Mode Shift, Preference Change, and Effect Persistence

Mon, 01 Sep 2025 00:00:00 GMT

Behavioral Responses to Congestion Pricing in New York City: Mode Shift, Preference Change, and Effect Persistence Shen, ChenAn This thesis examines the behavioral impacts of New York City’s congestion pricing policy on weekday peak-hour travel into the pricing zone. Using a two-stage Bayesian Multinomial Logit framework applied to monthly aggregate mobility data, the study disentangles underlying preference shifts from observed mode share changes in response to the toll. Stage 1 estimates population-level travel sensitivities to cost and time, while Stage 2 uses a hierarchical structure to capture heterogeneity across demographic segments defined by income, age, and gender. The analysis spans January–June 2025 and compares results to the same months in 2024 as a counterfactual scenario without pricing. Findings show that while the policy generated a sustained mode shift away from private automobiles toward public transit, preference adaptation varied by demographic group and evolved over time. Some cohorts reinforced the intended policy effects through reduced transit travel time sensitivity, while others exhibited partial reversal as cost sensitivity shifted. These dynamic patterns underscore the importance of evaluating both immediate and evolving behavioral responses when designing congestion pricing strategies and highlight the value of aggregate behavioral modeling for timely, data-driven policy assessment.

The Inhabited Arctic: Architecture, Time, and the Making of the Past in the Bering Strait (1760–1980)

Mon, 01 Sep 2025 00:00:00 GMT

The Inhabited Arctic: Architecture, Time, and the Making of the Past in the Bering Strait (1760–1980) Springstubb, Phoebe Our view of antiquity is not objective. From the eighteenth century on, the same actors and institutions involved in colonizing the Arctic shaped understandings of its deep past. Commercial whalers erected outposts on the Arctic Ocean’s edges; miners stripped tundra; trading companies raised forts. The demands of these projects complicated the Western imperial fiction of an Arctic without a past. Grappling with Arctic terrain, foreigners were confronted by a landscape inhabited not only by people and animals but by time and temporal imaginations that long preceded European colonization. They encountered contemporary Indigenous settlements coexisting with ancestral houses, fossil animals, the ruins of earlier colonial ventures, and ancient routes of exchange. This dissertation, centered on the Bering Sea and its adjacent geographies of eastern Siberia and Arctic North America, tells the story of how imperial upheaval and the rooting of colonial projects in the ground sparked a deliberate historiographic project to write the Arctic’s deep past. At the heart of this project was a conflict of different cultural views of time. Who had the right to narrate history in these northernmost borderlands? In episodes spanning two centuries, from the Russian empire’s claim to the Bering Sea to the rise of modern decolonial movements, this dissertation traces the central role of diverse Native architectures and technologies. Iñupiaq houses built from great whale skeletons, Unangax watercraft hewn from circulating driftwood, and Chukchi ice cellars carved into permafrost were both prisms for temporal explanations and sites driving change. Russian colonial administrators, British geologists, US ethnographers, Orthodox priests, and Soviet engineers co-opted them to the lineal, geological, eschatological, and paleolithic time that scaffolded imperial projects. Simultaneously, these material practices were vital sites for reinvention and identity, where Native nations built futures out of rupture. Illuminating how the ecological and epistemic limits to empire-building spurred new theories of Arctic time, this project shows history-making to be a crucial tool different states adopted to justify and naturalize their possessions of Native lands. At stake was not static historical truth but how politically situated temporalities structured their present-day actions. The ethical dimensions of deep time, imagined from the Bering Strait’s modern lands and seas, empowered empire’s practical work. How the past was conceived in different intellectual traditions informed whether animals and plants were exploitable resources or ancestors giving their bodies to architecture. This project contends that how people understood themselves as being in time was a decisive fulcrum ordering collective beliefs in what was owed to a larger, nonhuman world. Taking time as an analytical lens, this dissertation identifies repeated efforts to cleave the Arctic’s human history from nature’s past. Used to justify a wide range of colonial hierarchies and violence in the long nineteenth century, it underlies a contemporary bias toward seeing the Arctic as a region of deep naturalism. Viewed as a place where an “extreme” climate dominates manifold other historicities, the past so circumscribed continues to shape future possibilities.

Asset Limits, Savings Behavior, and Welfare: Evidence from the SIPP and a Life-Cycle Model

Mon, 01 Sep 2025 00:00:00 GMT

Asset Limits, Savings Behavior, and Welfare: Evidence from the SIPP and a Life-Cycle Model Gamble IV, James Monroe This paper examines how asset limits in means-tested welfare programs shape household saving behavior. I exploit cross-state variation in Temporary Assistance for Needy Families (TANF) asset limits by linking these limits to individual-level data from the Survey of Income and Program Participation (SIPP) and estimating ordinary least squares (OLS) regressions with state and year fixed effects. I find that a $1 increase in the liquid asset limit corresponds to a $0.75 decrease in non-housing wealth among single mothers without a high school diploma. This suggests that less stringent asset tests reduce incentives to save, consistent with models in which more generous public insurance lowers the need for precautionary saving. To interpret these findings, I develop a dynamic life-cycle model of saving under income and medical expense risk, calibrated to key moments from the Hubbard, Skinner, and Zeldes framework. The model embeds Medicaid-style transfer rules and a guaranteed consumption floor. Simulations indicate that a $7,000 consumption floor can reduce median assets by up to 20% among low-education households, reflecting a decrease in self-insurance as public support increases. I then extend the model to include Achieving a Better Life Experience (ABLE) accounts, which are tax-advantaged savings vehicles for individuals with disabilities exempt from means testing. Simulations indicate that ABLE eligibility increases early-life consumption by approximately $10,000 and reduces retirement savings, with account holders shifting more spending into their working years. Together, these results yield a direct mapping from policy levers, including asset-limit generosity, earnings disregards, childcare subsidies, and ABLE exemption rules, to predicted shifts in median household assets. This offers policymakers a practical tool to balance public insurance and private precautionary savings.

Building the 3D Genome from the Ground Up: Local Interactions Give Rise to Global Order

Mon, 01 Sep 2025 00:00:00 GMT

Building the 3D Genome from the Ground Up: Local Interactions Give Rise to Global Order Athreya, Advait The three-dimensional organization of the genome within the nucleus plays a central role in determining gene regulation and establishing cellular identity, but the mechanisms by which local molecular interactions give rise to global chromatin architecture remain an active area of study. Interactions between nucleosomes—modulated by histone tail post-translational modifications, histone sequence variants, and the DNA sequence itself—are thought to be a major driver of this emergent structure. In this thesis, I address the question of how these intrinsic physicochemical properties of nucleosomes drive the formation of large-scale structures such as chromatin compartments. I develop a theoretical framework based on Flory-Huggins solution theory to derive pairwise internucleosome contact energies from the results of condense-seq, a novel experimental technique that measures the phase separation likelihood of native nucleosomes. I then use these derived energies to parameterize coarse-grained molecular dynamics simulations of chromatin at various resolutions, ranging from 25kb segments to simulate an entire chromosome, down to individual nucleosomes to simulate up to 10Mb genomic regions. These simulations demonstrate that the intrinsic nucleosome properties alone can capture a significant degree of A/B compartment formation observed in Hi-C experiments, despite the deliberate exclusion of all other factors such as loop extrusion and transcriptionfactor-mediated phenomena. This finding establishes that local nucleosome properties play a fundamental role in genome organization. To capture more detailed chromatin physics, I develop an extended chromatin force-field that incorporates anisotropic nucleosome stacking interactions and linker DNA properties using a novel approach for simulating reversible bond formation in molecular dynamics. This model reveals how nucleosome stacking strength, linker DNA geometry, and torsional stress collectively influence higher-order structures. Early results show that the linker-length-dependent DNA torsion contributes to nematic ordering of chromatin, consistent with experimental studies. Future development of this model will enable probing of discrete domain formation observed in imaging studies. Finally, I address a critical consideration for researchers in the chromatin organization field when analyzing Hi-C results. I compare two software tools — cooltools and dcHiC — highlighting the importance of careful parameter selection and analytical choices in designing workflows to ensure reproducible research. Taken together, this work establishes a quantitative, bottom-up modeling framework that directly links the local physicochemical properties of nucleosomes to the global principles governing three-dimensional genome organization. It provides a complementary approach to more data-driven top-down models that have made significant inroads but are challenging to interpret mechanistically. With further development, the work presented in this thesis will contribute towards predicting the structural consequences of specific epigenetic modifications and move us closer to understanding the molecular grammar of chromatin and its role in cellular function and disease.

On Subannual Variability in the Abyssal Ocean

Mon, 01 Sep 2025 00:00:00 GMT

On Subannual Variability in the Abyssal Ocean Chen, Si Yuan The abyssal ocean is a critical yet understudied component of the climate system and is of growing economic interest. This thesis combines field observations and numerical modeling to advance our understanding of subannual variability in the abyssal ocean and its broader implications. First, hydrographic measurements from the Clarion-Clipperton Zone of the tropical Northeastern Pacific are used to characterize the structure and variability of the bottom mixed layer (BML) in a region targeted for deep-sea mining. The observations reveal a spatially and temporally variable BML with a mean thickness of ~250 m and influenced by interactions with mesoscale eddies and abyssal thermal fronts. A simplified model of sediment transport suggests that such variations in BML structure could significantly influence the dispersal of sediments resuspended by seabed mining activities. Second, idealized model experiments are conducted to explore the genesis of benthic storms – episodes of strong near-bottom flows and sediment entrainment – underneath an unstable, surface-intensified jet resembling the Gulf Stream east of Cape Hatteras. In these experiments, the baroclinic instability of the jet gives rise to deep cyclonic and anticyclonic eddies through eddy barotropization and to high levels of eddy kinetic energy at abyssal depths through the convergence of vertical eddy pressure fluxes. The near-bottom currents are comparable in magnitude to those observed during benthic storms, with vertical shears strong enough to produce BMLs up to O(100) m thick. Deep cyclonic eddies transport particles from near the bottom over the entire BML and could contribute to benthic nepheloid layers. The results suggest that the abyssal response to the intrinsic instability of surface-intensified currents could contribute significantly to subannual variability near the seafloor. Third, a model simulation of western North Atlantic circulation is performed to study the deep cyclones (DCs) observed beneath Gulf Stream meander troughs. The characteristics of the simulated DCs compare well with field observations. The negative pressure tendency during cyclogenesis arises from a small imbalance between the sea surface depression and the vertically-integrated increase in seawater density. Vortex stretching is the primary source of cyclonic vorticity, while vortex tilting is a non-negligible sink. The deep pressure tendency, vorticity fluxes, and ageostrophic flows are diagnosed, and their similarities and differences with mid-latitude synoptic cyclones in the atmosphere are discussed. Near-bottom currents in DCs dominate the basin-scale bottom energy dissipation and transport fluid over ≥1000 km horizontally and O(100) m vertically within 3~4 months, suggesting that they provide an efficient mechanism for tracer and material transport in the abyssal interior. Collectively, this thesis highlights the importance of transient, mesoscale processes in contributing to subannual variability in the abyssal ocean, particularly near the seafloor. The findings have broader relevance for monitoring the environmental impacts of human activities, including deep-sea mining and carbon sequestration. While further questions remain for future investigation, this work underscores the need for sustained in-situ observations in the abyssal ocean and calls for the implementation of high vertical resolution in numerical ocean circulation models.

Democratizing High-Performance DSL Development with the BuildIt Framework

Mon, 01 Sep 2025 00:00:00 GMT

Democratizing High-Performance DSL Development with the BuildIt Framework Brahmakshatriya, Ajay Modern high-performance software from a variety of domains relies on hand-written and hand-optimized libraries to obtain the best performance. Besides general fine-grained operators that can be composed to write entire applications, these libraries also provide coarser-grained fused and hand-optimized operators that are much faster due to being optimized for a specific sequence of operations. However, as application needs keep growing, library writers are not able to keep up and have to make the tradeoff of either sacrificing performance or generality. Domain-specific languages or DSLs are able to break this tradeoff by automatically generating the best implementation for any arbitrary sequence of operations specified by the end user. However, DSL compilers suffer from a bigger challenge that they require a lot of compiler knowledge to implement parsers, IR, analysis and transformations, and code generation, which is outside the scope of a typical domain expert. To make compiler technology and the benefits of code-generation more accessible to domain experts, I propose the use of multi-stage programming to allow writers to write library-like code while also combining it to generate the most efficient implementation for any whole program. In this thesis, I discuss the design of different multi-stage programming systems, the benefits and drawbacks. Next, I propose Re-Execution Based Multi-Staging (REMS) that addresses a critical flaw in many imperative Multi-Staging systems - the side-effect leak problem. I introduce BuildIt, an implementation of REMS in one of the most popular languages for writing high-performance applications, C++ in a type-based, lightweight way without changing the compiler. I describe the internals of BuildIt and how it implements the key features of REMS. Furthermore, I describe a set of extensions implemented on top of BuildIt that facilitate the development of high-performance DSLs with ease. I show the application of BuildIt to create three DSLs - EasyGraphit, NetBlocks, and BREeze that target graph analytics, ad-hoc network protocol generation, and Regex matching. All these case studies show 10-100x reduction in the amount of effort required to implement these DSLs that perform on-par with or better than state-of-the-art compiler frameworks while targeting diverse architectures like CPUs and GPUS. Finally, I introduce D2X, a system that is designed to add extensible and contextual debugging support to DSL implementations without having to make any changes to off-the-shelf debuggers or mess with complex debugging formats. Next, I show how applying D2X to the BuildIt system greatly improves the debugging experience for all DSLs written with BuildIt.

Tailoring Li₄Ti₅O₁₂ Thin Film Carrier Kinetics Through Solid Solution Doping for Battery and Memristor Applications

Mon, 01 Sep 2025 00:00:00 GMT

Tailoring Li₄Ti₅O₁₂ Thin Film Carrier Kinetics Through Solid Solution Doping for Battery and Memristor Applications Buzzell, Drew E. A Lithium titanate, Li₄Ti₅O₁₂ (LTO4), due to its zero-strain behavior during cycling, excellent chemical stability and cyclability, is a promising anode material for solid-state batteries (SSB) applications. As a thin film, its applications expand to integrated circuits, sensors, flexible batteries, IoT devices, and memristors. Across these, precise control of mixed Li⁺ ionic–electronic transport is vital. While dopants have been shown to improve electron conduction and Li⁺ diffusion in LTO4 powders, thin-film studies remain limited. To bridge this gap, we investigate solid solution dopants (Nb⁵⁺, V⁵⁺, Mg²⁺, Cu²⁺) and their effects on LTO4 thin-film kinetics and performance in batteries and memristors. Films doped with Mg, Cu, Nb, and V with a 0.2M dopant concentration were deposited on Nbdoped SrTiO₂ substrates. Cyclic voltammetry and impedance spectroscopy show that Mg, Nb, and V improve kinetic metrics, while Cu reduces diffusivity but boosts electronic conductivity. Through galvanostatic cycling-based capacity, rate capability, and stability measurements, we found that while all dopants displayed enhanced rate performance, the capacity improved only with Mg, Nb, and V. Furthermore, the Mg-doped film was found to have an unstable capacity leading to Nb- and V-doped thin-films as the best overall performing battery anodes. For memristors, current–voltage cycling measurements revealed that low concentrations (0.05 M) of Cu and Nb doped devices presented the largest improvements in cycle-to-cycle stability, switching ON-voltages, ON-OFF current ratios, and lower loss in peak current with increasing scan rate. With increasing dopant concentrations however, devices would see relative drops in performance. In summary, the inclusion of dopants in LTO4 at the right concentration level leads to improvements in both battery and memristor performance allowing for one material multi-functional systems.

Strain-resolved transcriptomics: exploring functional heterogeneity of the gut microbiota in health and disease

Mon, 01 Sep 2025 00:00:00 GMT

Strain-resolved transcriptomics: exploring functional heterogeneity of the gut microbiota in health and disease Burgos Robles, Emanuel Felipe The gut microbiome plays a critical role in inflammatory bowel diseases (IBDs), yet current analyses treat bacterial species as functionally uniform, ignoring extensive strain-level diversity that may drive disease mechanisms. Here, we developed a strain-resolved metatranscriptomics framework to investigate how transcriptional activity varies across bacterial lineages and relates to IBD pathogenesis. Using paired metagenomics and metatranscriptomics data from 1,067 fecal samples (103 IBD and 335 non-IBD patients), we first constructed phylogenetic trees for over 250 bacterial species using the single nucleotide variants within essential housekeeping genes, enabling the identification of bacterial strains. Next, we devised a statistical approach to assign mRNA reads to these strains, leveraging the natural genetic variation that is present across them. My analysis revealed that closely related bacterial strains exhibit dramatically different transcriptional programs, with some strains enriched in IBD patients showing upregulation of genes involved in stress response, sugar metabolism pathways, and antimicrobial resistance. Notably, we identified transcriptionally active but genomically low-abundance taxa, highlighting the importance of measuring the transcriptional activities of strains beyond species composition. Lineage-aware differential expression analysis uncovered strain-specific adaptations to inflammatory environments. This strain-resolved approach provides a powerful framework for understanding microbial functional heterogeneity and identifying specific bacterial lineages that could potentially contribute to disease pathogenesis, potentially guiding more targeted microbiome-based therapeutic interventions.

The Role of CH–π Interactions in Protein-Carbohydrate Binding

Mon, 01 Sep 2025 00:00:00 GMT

The Role of CH–π Interactions in Protein-Carbohydrate Binding Keys, Allison M. Protein-carbohydrate binding is essential for biological processes, including cellular recognition and immune signaling. Binding is driven by several types of non-covalent interactions: hydrogen bonding, metal ion coordination, and the less well-understood CH–π interactions. CH–π interactions are pervasive in protein-carbohydrate binding sites and have emerged as critical drivers of protein–carbohydrate recognition; however, the energetics of CH–π stacking interactions, their orientational landscapes, and their interplay with other non-covalent interactions have been unclear. In this thesis, I identified carbohydrate-aromatic CH–π stacking interactions from crystallographic structures in the Protein Data Bank. I performed quantum mechanical calculations to quantify interaction energies and found that CH–π stacking interactions can be more favorable than hydrogen bonds. Using atomistic simulations, I also demonstrated that CH–π stacking interactions are necessary for human galectin-3 binding to lactose. To assess the orientational landscape of CH–π stacking interactions, I evaluated the orientations of CH–π stacking interactions formed by β-D-galactose and found that numerous orientations are highly favorable. I then identified carbon atom distances that define an orientational landscape for these interactions. To assess the interplay between non-covalent interactions in protein-carbohydrate binding sites, I used CH–π distance features to bias metadynamics simulations of a curated set of protein–β-D-galactoside complexes. From these simulations, I found that while bound carbohydrates sample many CH–π stacking orientations, the hydrogen bonds in the protein binding site drive the optimal orientation of each ligand. Longer carbohydrate ligands with more hydrogen bonding constraints have more specific orientational dependence, while ligands in binding sites with a reduced number of hydrogen bonds occupy a broader range of orientations. Unlike hydrogen bonds, CH–π stacking interactions confer orientational flexibility: enzymes can exploit multiple CH–π stacking interactions to facilitate the translocation of polysaccharide substrates. Extending this analysis to other carbohydrates, I showed that carbohydrate stereochemistry drives the orientational preferences of CH–π stacking interactions; however, there is also a tradeoff between the presence of hydrogen bonds to charged amino acids and the CH–π interaction strength for each carbohydrate. Overall, this thesis demonstrates that CH–π interactions are favorable and confer high orientational flexibility and that hydrogen bonds act in concert with CH–π interactions to stabilize protein-carbohydrate binding. Tuning the number and positions of these interactions through protein engineering should alter protein selectivity and ligand movement in protein binding sites.

Quantum Networking using Waveguide Quantum Electrodynamics

Mon, 01 Sep 2025 00:00:00 GMT

Quantum Networking using Waveguide Quantum Electrodynamics Almanakly, Aziza The architectural principle of modularity enables the construction of complex systems from simpler components, each responsible for a particular function. The quantum computer is an intricate system comprising fragile, error-prone parts known as qubits. Entanglement distribution across a network of non-local processing modules facilitates robust and extensible quantum computation. In modular quantum architectures, photons are natural quantum information carriers which propagate through interconnects between processing nodes. In this thesis, we engineer a quantum interconnect between superconducting modules underpinned by the physics of waveguide Quantum Electrodynamics (wQED). First, we realize a multi-qubit module that exploits quantum interference to emit microwave photons into a waveguide with a specified propagation direction. Next, we construct the quantum interconnect by coupling two modules to a common waveguide and demonstrate directional (chiral) photon emission and absorption. Finally, using this chiral quantum interconnect, we generate remote entanglement, establishing a key resource for distributed quantum computation in an all-to-all network architecture.

Large Language Models and Quantifying the Regulatory Expenses of Affordable Housing: A Thorough Examination Utilizing Generative Assessment

Mon, 01 Sep 2025 00:00:00 GMT

Large Language Models and Quantifying the Regulatory Expenses of Affordable Housing: A Thorough Examination Utilizing Generative Assessment Xu, Bangjie This thesis presents an innovative methodology using Large Language Model-based methods to extract and quantify housing regulations from municipal zoning codes, making possible the most comprehensive examination of regulatory costs at the municipal level across California to date. A multi-staged extraction framework is devised that delivers 85-95% accuracy in the identification and standardization of complex regulatory requirements from legal documents. Applying this methodology to over twenty California cities over the period 2015-2025, it is estimated that regulatory constraints raise the cost of developing a housing unit by roughly between 5% to 10% (or $50,000 and $100,000+) per housing unit, with the most acute constraints in the state’s coastal metros. This method is used to find that factors such as regulation costs limit housing supply elasticity from 1.24 in low-regulation jurisdictions to 0.08 in high-regulation areas. The LLM-based framework allows us to conduct analyses at an unprecedented scale and granularity and to reveal, for example, that the relaxation of regulation by streamlining policies like the Los Angeles Transit Oriented Communities program boosts housing production in eligible zoned areas by 43%. This study makes significant contributions to the restructuring of California’s housing regulation system in response to the affordability crisis, and its methodology presents a replicable tool for regulatory analysis in other policy domains.

Essays in Macro-Finance

Mon, 01 Sep 2025 00:00:00 GMT

Essays in Macro-Finance Batista, Quentin In Chapter 1 (joint with J.R. Scott), we revisit the high-frequency and narrative approaches to estimating the effects of monetary policy shocks. We find that state-of-the-art estimates using both approaches are biased: high-frequency estimates due to nonlinear predictability and narrative estimates due to regularization. To correct for the bias in these approaches, we propose a new estimation procedure called LP-DML that combines ideas from double/debiased machine learning with the local projections framework. We find that LP-DML results in significantly smaller effects of monetary policy on macroeconomic outcomes. In Chapter 2 (joint with Taisuke Nakata and Takeki Sunakawa), we study the following question: how a central bank credibly implement a ”lower-for-longer” strategy? To answer this question, we analyze a series of optimal sustainable policy problems—indexed by the duration of reputational loss—in a sticky-price model with an effective lower bound (ELB) constraint on nominal interest rates. We find that, even when the central bank lacks commitment, the central bank can still credibly keep the policy rate at the ELB for an extended period though not as extended as under the optimal commitment policy—and meaningfully mitigate the adverse effects of the ELB constraint on economic activity. In Chapter 3, I examine the impact of central bank real estate purchases on financial markets, focusing on the Bank of Japan’s (BoJ) intervention in the Real Estate Investment Trust (REIT) market. Using a regression discontinuity design that exploits a discontinuity in the BoJ’s policy rule, I find that a typical intervention — amounting to about 0.014% of market capitalization — leads to an increase of 0.1% to 0.2% of REIT prices in the hours following the intervention. However, at longer horizons, the interventions do not have a significant effect on REIT prices. These findings suggest that the BoJ did not achieve the program’s intended objective of significantly reducing the risk premium on real estate assets.

Toward an Integrative Study of Human-AI Interaction

Mon, 01 Sep 2025 00:00:00 GMT

Toward an Integrative Study of Human-AI Interaction Alsobay, Mohammed As artificial intelligence (AI) systems are increasingly embedded in the workflows of individuals and groups, designers and researchers of human-AI interaction (HAI) navigate a vast design space of possible configurations, making decisions that span algorithmic parameters, interface choice, and interaction protocols. This thesis develops an integrative approach that examines how design factors combine and interact to determine the outcomes of human-AI collaboration. Chapter 1 synthesizes prior HAI research into a coherent design space framework encompassing algorithms, interfaces, users, and task settings, motivating a research program for systematic exploration of interdependencies between these factors. Chapters 2 and 3 turn to group-AI interaction through large-scale behavioral experiments. Chapter 2 investigates how social information---both direct conversation and peer behavior indicators---affects individual reliance on algorithmic decision support. The study reveals that while social information modulates the effects of performance feedback and model explanations on reliance, it does not improve predictive accuracy, illuminating critical tensions between social mechanisms and system design. Chapter 3 examines large language models as facilitators of group deliberation in hidden profile tasks. While LLM facilitation increased information sharing volume, density, and breadth, it did not improve decision quality, highlighting fundamental challenges in group-AI system design beyond information aggregation. Chapter 4 advances an integrative approach to HAI research, emphasizing shared design spaces, systematic exploration strategies, and predictive models that generalize across contexts. The chapter provides methodological guidance and a tractable roadmap for advancing this integrative research agenda, laying the foundation for a more context-aware science of human-AI collaboration.

Leveraging design to build with less: Evaluating the embodied carbon reduction potential of architectural design across scales

Mon, 01 Sep 2025 00:00:00 GMT

Leveraging design to build with less: Evaluating the embodied carbon reduction potential of architectural design across scales Feickert, Kiley Reducing embodied carbon (EC) in structural systems -- the most significant contributor to EC in a building -- is urgent to address the simultaneous need to reduce global warming and increase urban density. Much of the policy and research to date to reduce EC has focused on material-scale interventions or substitutions. However, EC depends on both: 1) the carbon intensity of the processes used to manufacture construction materials, and 2) the volume of raw materials required. Architects have significant agency to reduce the volume of structural materials in a building (and the resulting emissions) since the required quantity depends on design decisions architects make, including column spacing, structural typology, massing, etc. To date, most methods used to estimate EC during early-stage design do not: 1) integrate with architects’ existing design workflows, 2) evaluate multiple material systems simultaneously, and/or 3) include structural analysis to estimate material quantities. This functionality is critical so that designers can understand which decisions EC is sensitive to and evaluate design and EC tradeoffs before significant carbon is locked in. To address this problem, this dissertation presents a method towards transparent estimation of structural material quantities, intending to inform architectural design and policy, or other emerging EC standards. This method is used to contribute an analysis of the effectiveness of emerging U.S. EC policies, which focus on different scales of intervention, at the building scale. These policies are evaluated in isolation and in combination with strategic design levers that take advantage of structural mechanics to reduce material quantities for various building configurations and material systems. It finds that the most prominent policy approach, “Buy Clean” materials, only reduces EC by ~9% and ~16% for steel and concrete systems, respectively, compared to strategic design choices that have the potential to yield savings of up to ~79%. This dissertation also identifies building massing as a key lever in the EC outcomes of structural systems and proposes a method to quantify the impact of massing using automated structural design and analysis. It finds that in some situations, cantilevered massing typologies can be materialized for no carbon penalty if efficient configurations are used. Indeed, if inefficient configurations are used, they can incur a significant carbon penalty (2.4x) compared to normative massing. The presented results highlight the potential of design to reduce demand-side EC across scales.

Generalizable Robot Manipulation through Unified Perception, Policy Learning, and Planning

Mon, 01 Sep 2025 00:00:00 GMT

Generalizable Robot Manipulation through Unified Perception, Policy Learning, and Planning Fang, Xiaolin Advancing robotic manipulation to achieve generalization across diverse goals, environments, and embodiments is a critical challenge in robotics research. While the availability of data and large-scale training has brought exciting progress in robotics manipulation, current methods often struggle with generalizing to unseen, unstructured environments and solving long-horizon tasks. In this thesis, I will present my work in robot learning and planning that enables multi-step manipulation in partially observable environments, towards general-purpose embodied agents. Specifically, I will talk about my work in 1) constructing a modular framework that estimates affordances with learned perception models with task-and-motion-planning (TAMP) for object rearrangement in unstructured scenes, 2) learning generative diffusion models of robot skills, which can be composed to solve unseen combination of environmental constraints through infeference-time optimization, 3) leveraging large vision-language models (VLMs) in building task-oriented visual abstractions, allowing skills to generalize across different environments with only 5 to 10 demonstrations. Together, these approaches contribute to the generality and scalability of embodied agents towards solving real-world manipulation in unstructured environments.

Towards High-Dimensional Generalization in Neural Networks

Mon, 01 Sep 2025 00:00:00 GMT

Towards High-Dimensional Generalization in Neural Networks Boopathy, Akhilan Neural networks excel in a wide range of applications due to their ability to generalize beyond training data. However, their performance degrades on high-dimensional tasks without large-scale data, a challenge known as the curse of dimensionality. This thesis addresses this limitation by pursuing three key objectives aimed at understanding and improving neural network generalization. 1. We aim to investigate the scaling laws underlying generalization in neural networks including double descent, a phenomenon in which as a model’s capacity or training data is increased, the test error temporarily increases at a certain point before continuing to decrease. In particular, we will have two goals: 1) a better understanding of when double descent can and cannot be empirically observed and 2) a better understanding of scaling laws with respect to training time. 2. Inductive bias refers to the set of assumptions a learning algorithm makes to predict outputs on inputs it has not encountered. We propose quantifying the amount of inductive bias required for a model to generalize well with a fixed amount of training data. By developing methods to measure inductive bias, we can assess how much information model designers need to incorporate into neural networks to improve their generalizability. This quantification can guide the design of harder tasks that better test a model’s generalization. 3. Finally, we aim to develop new methods to enhance neural network generalization, particularly focusing on reducing the exponential number of training samples required for high-dimensional tasks. This involves creating algorithms and architectures that can learn effectively from limited data by incorporating stronger inductive biases. In particular, we will focus on two inductive biases in particular: 1) learning features of the training loss landscape correlated with generalization and 2) using modular neural network architectures. We expect that these techniques can improve generalization, particularly in high-dimensional tasks. Together, these contributions aim to deepen our theoretical understanding and develop practical tools for enabling neural networks to generalize effectively from limited data.

Remote Control: Art, Technology, and the Politics of Distance (1966-1972)

Mon, 01 Sep 2025 00:00:00 GMT

Remote Control: Art, Technology, and the Politics of Distance (1966-1972) Wexelblatt, Nina Rrose Platforms carrying dancers across a stage, doors sliding open as if by magic, and simultaneous Happenings in Berlin and Buenos Aires: remote control promised thrills as postwar artists experimented with technologies of distance. Focused on the half-decade between 1966 and 1972, this thesis intervenes in the history of art and technology to argue that a desire to activate the supposedly empty space between artist, art object, and audience effected a new fixation on the nature of that distanced interval, leading artists to incorporate actual remote control technologies into their work. This impulse grew from an unorthodox reading of the work of modernist painters, particularly Jackson Pollock. Where a generation of critics had canonized “presentness” and medium specificity, a younger cohort read the work differently, finding in it permission to embrace remoteness, intermedia experimentation, and political messaging. Artists including Robert Rauschenberg, Allan Kaprow, Marta Minujín, Wolf Vostell, and Carolee Schneemann, among others, undertook radical experiments with remote systems, often in collaboration with engineers. Theirs was not a technocratically neutral position; this thesis demonstrates that these artists consciously cast the “remoteness” enabled by new technologies as a charged concept, just as controlled distance emerged to define military and industrial relations on domestic, urban, and geopolitical scales. Remote control enabled artists to incorporate, not reject, the expanding frames of reference taking place outside of the sanctioned spaces of the art studio or gallery, from automation to satellite communications to warfare. Artists’ uses of remote technologies intentionally surfaced questions about critical power relations, tying the stakes of their work to debates about the future of U.S. social and economic control and development. In doing so, it also crystallized a newly diffuse, participatory artistic subject: the controller. The introduction theorizes “remote control” in historical and historiographic context. A second chapter follows Automation House (1970-1972), a Manhattan art space that combined labor mediation and media art to experiment with the American postindustrial labor economy to come. A third chapter centers on Three Country Happening (1966), which took place in New York, Buenos Aires, and Berlin, supposedly mediated by satellite—foiled by the uneven development of the Cold War-era satellite system itself. A fourth chapter delves into Snows (1967), a multimedia performance in protest of the war in Vietnam, which incorporated audience-controlled feedback sensors. A concluding discussion traces the ongoing nature of remote control as it implicates artists and audiences alike in a network of shared responsibility.

Reimagining Public Land: Municipal Land Use to Ease Housing Crisis in Boston

Mon, 01 Sep 2025 00:00:00 GMT

Reimagining Public Land: Municipal Land Use to Ease Housing Crisis in Boston Murphy, Ryan Boston is in the midst of a severe housing crisis, driven by decades of underproduction, rising construction costs, restrictive zoning, and an inelastic real estate market that has resulted in persistent affordability challenges. This thesis explores the untapped potential of city-owned land as a powerful tool to increase housing supply and affordability in Boston. Using Boston’s 2022 Citywide Land Audit and detailed development assumptions, the analysis estimates that between 19,000 and 31,000 new housing units could be constructed across city-controlled parcels, including between 3,200 and 6,100 affordable units under the current Inclusionary Development Policy. The research draws on case studies from peer cities such as Chicago and Atlanta where municipal land has been successfully leveraged through transparent disposition processes, fast-tracked entitlements, and flexible affordability models. It argues for a policy shift in Boston toward a more streamlined, market-aware, and scalable land release strategy that prioritizes speed, cross-subsidization, and financial feasibility. Key recommendations include expanding the Welcome Home, Boston program to include mixed-income and rental housing, implementing predictable RFP cycles, offering tax abatements, and expediting the entitlement process.

Modular Zipping for Transformable and Dynamic Systems

Mon, 01 Sep 2025 00:00:00 GMT

Modular Zipping for Transformable and Dynamic Systems Hagemann, Niklas There is a need for products, machines and environments that can change shape, transform and evolve according to their use. This thesis proposes the design of a simple, modular actuator based on reversible folding and interlocking (zipping) of flexible 3D printed strips. The proposed zipper design allows for continuous control states between a compact and fully deployed state. The modular actuators can be integrated into a variety of systems to enable compact, shape- and stiffness-changing structures, robots and other devices. Designs are presented for single- and double-zipper modules using the same basic zipper design. The modules can be used as modular components of compact robotic systems with the ability to expand and contract according to their environment, or used as adjustable structural components to create deployable, shape-and stiffness-changing objects. The zipper design points the way towards simplified mono-material components that embed transformation and reversibility into everyday devices, products and spaces, and enabling objects that are as easy to transform, reconfigure and reverse as they are to manufacture.

Embodied Representation of Time in Virtual Reality

Mon, 01 Sep 2025 00:00:00 GMT

Embodied Representation of Time in Virtual Reality Kim, Suwan Recent advancements in 3D graphics and AI-assisted generative techniques have accelerated the creation of realistic scenes for immersive technologies, including virtual reality, yet most systems continue to encode time as a linear parameter, relying on timeline-based playback. Mesh-based representations are typically constrained by fixed topologies and rely on predefined animations, which limit their capacity to encode temporal change as a spatial or perceptual phenomenon. In reality, human experience of time is embodied and dynamic, perceived through interaction and memory. Existing digital systems fail to capture this dimension, reducing time to a passive parameter. This thesis proposes a framework for representing time as an embodied and spatial dimension within virtual reality by embedding it directly into the geometry and interaction logic of point cloud data. The system consists of three parts: (1) processing 2D images into layered volumetric point clouds to enable structural fluidity and temporally responsive spatial form; (2) enabling perceptual and spatial modulation in response to user distance and contact, with color influencing the character of change and opacity shaping its perceptual reveal at both global and local scales; and (3) enabling real-time visualization of modulated point cloud through a custom pipeline optimized for mobile virtual reality. By embedding temporal dynamics directly into geometry and interaction logic, this thesis contributes a novel representational approach to spatiotemporal modeling in immersive systems. By doing so, we create new opportunities for architectural visualization, interactive simulations, game design, and reimagining how we perceive and construct digital spaces.

Essays on Behavioral Economics and Sophisticated Procrastination

Mon, 01 Sep 2025 00:00:00 GMT

Essays on Behavioral Economics and Sophisticated Procrastination Chen, Xi Procrastination is a widespread yet complex behavior that resists simple explanation. This dissertation integrates theoretical modeling with experimental evidence to examine procrastination through the lens of sophisticated decision-making. It reframes procrastination not merely as a deviation from rationality, but as a behavior shaped by strategic trade-offs, self-awareness, and individual heterogeneity. The first essay develops a theoretical model of Perfectionistic Procrastination, proposing that individuals with high internal standards may delay tasks not as a simple lapse in self-control, but as a strategic response to the anticipated costs of sustained effort. In this framework, deadlines act as external constraints that help perfectionists limit open-ended striving and bring tasks to completion. An accompanying experiment tests the model’s prediction and finds that perfectionists are more likely to prefer deadlines. These results suggest that, in some cases, procrastination may reflect a structured strategy rather than a purely irrational failure of self-control. The second essay explores the phenomenon of Sophisticated Procrastination, challenging traditional models that attribute procrastination to naïveté. Instead, it proposes that even individuals who are aware of their tendency to delay may struggle to act on that awareness. Two experimental studies using a menu-choice framework examine how people choose task timings. In Study 1, participants preferred earlier deadlines when flexibility was available but shifted toward later options when required to commit, revealing a gap between intention and action. Study 2 identified diverse patterns of deadline preferences: while many participants actively avoided the latest possible deadline, their hesitation to commit to any specific deadline suggests a deeper tension rooted in uncertainty or discomfort with commitment. These findings provide early empirical support for Sophisticated Procrastination, indicating that self-awareness alone may not be sufficient to overcome procrastination. The third essay introduces the idea of Prosocial Procrastination, describing the tendency to delay tasks that benefit others, such as charitable activities, more than those with self-interested outcomes. Using two distinct experimental designs, one based on conjoint analysis and the other on single-attribute choice, the studies show that individuals are more likely to prefer longer deadlines when working for a charity than when working for themselves. These findings offer suggestive evidence for Prosocial Procrastination and contribute to the growing literature on the intersection of social preferences and time preferences.

Patent Visibility and the Diffusion of Trapped Knowledge: Evidence from US Grants

Mon, 01 Sep 2025 00:00:00 GMT

Patent Visibility and the Diffusion of Trapped Knowledge: Evidence from US Grants Yao, Randol H. Valuable knowledge developed in one part of the world may remain “trapped" locally due to frictions in how knowledge is recognized and shared globally. This paper examines how granting US patents to foreign-origin inventions—by elevating their visibility and credibility— untraps the knowledge and facilitates global diffusion. Using examiner leniency as an instrument, complemented by a difference-in-differences design, I find that US grants of home country patents significantly increase both the likelihood and intensity of forward citations, including marked increases from third countries. A novel measure of “trappedness” reveals that knowledge from historically more trapped countries and sectors sees larger diffusion benefits after the US grants. These findings highlight the central role of the US as a platform of global knowledge recognition and diffusion, particularly in turning overlooked ideas into globally relevant innovations.

Characterization of the East China Sea Continental Shelf Circulation Northeast of Taiwan Surrounding Mien-Hua Canyon

Mon, 01 Sep 2025 00:00:00 GMT

Characterization of the East China Sea Continental Shelf Circulation Northeast of Taiwan Surrounding Mien-Hua Canyon Rafferty, Lieutenant Commander Keefe Submarine canyons have a proven and direct influence on continental shelf circulation and flow dynamics, especially in relation to western boundary currents. There are two key circulation features northeast of Taiwan on the East China Sea continental shelf: (1) the cold dome, a cyclonic feature that appears primarily in summer and is associated with upwelling, and (2) Kuroshio intrusions onto the continental shelf in the vicinity of Mien-Hua Canyon. This paper is a descriptive physical oceanography study with a focus on characterizing the circulation patterns northeast of Taiwan surrounding Mien-Hua Canyon, closely correlating these patterns with the migration of the Kuroshio and its variability and intrusions onto the southern East China Sea continental shelf, leading to the formation of the cold dome. The Institute of Oceanography at the National Taiwan University and WHOI executed a joint international field survey at Mien-Hua Canyon aiming to improve the understanding of canyon flow dynamics between the East China Sea continental shelf northeast of Taiwan and the Kuroshio as the North Pacific Gyre westward boundary current. This joint oceanographic expedition expands on previous joint US/Taiwan physical oceanographic and ocean acoustic studies in the China Seas dating back to ASIAEX in the South China Sea during 2000-2001 and QPE in the East China Sea during 2008-2009. The strengthening and weakening of Kuroshio transport and intensity northeast of Taiwan is closely correlated to the timescales of mesoscale westward propagating eddies arriving to the East Taiwan Channel. When a canyon has a Rossby number ~1 or Rossby radius equivalent to the width of the canyon in a region of left-bounded flow, induced cyclonic flow will experience an upwelling regime within the canyon system with dominant upwelling located at the downstream canyon rim vertically constrained by Rossby Height. Observational analysis of canyon bottom-moored ADCPs and vertical temperature arrays supports previous theory on submarine canyon dynamics on a continental shelf. Satellite sea surface temperature and absolute dynamic topography observations render the formation of a cold dome northeast of Taiwan coincident with this joint oceanographic survey.

From Wallets to Wages: Consumer Income, Job Design, and Pay Disparities

Mon, 01 Sep 2025 00:00:00 GMT

From Wallets to Wages: Consumer Income, Job Design, and Pay Disparities Roh, Soohyun Pay differences between organizations are a key source of wage inequality. I propose a novel account of these differences by starting from the consumers that these businesses serve. Firms that serve high-income consumers specialize jobs into higher-paying and higher-skilled positions focused on quality, while those that serve lower-income consumers emphasize cost minimization by requiring workers to perform a wider range of general tasks. Matching consumer foot traffic data and establishment-level wage records, I find that establishments serving higher-income consumers pay their workers more. This effect holds comparing among establishments in the same neighborhoods and industries. Longitudinally, establishments increase wages when they shift toward higher-income customers. Analysis of online job postings further reveals that jobs at higher-income-serving firms involve a narrower set of tasks that command higher market value. These findings show how consumer markets shape firms’ internal job design and contribute to pay inequality across organizations.

The Fate of Federal Buildings: Real Estate Disposition and the Future of Washington, D.C.

Mon, 01 Sep 2025 00:00:00 GMT

The Fate of Federal Buildings: Real Estate Disposition and the Future of Washington, D.C. Mulcahy, Robby L. The United States federal government is the largest property owner in the country, with more than 370 million square feet of real estate under its control. Much of this portfolio is outdated, underutilized, and located in the urban cores of American cities. Nowhere is this more evident—or more consequential—than in Washington, D.C., where the federal government controls approximately 27% of the office market. As federal agencies adopt hybrid work models, and as the operational needs of government evolve, the existing real estate footprint has become increasingly inefficient, expensive, and misaligned with civic and market realities. This thesis investigates the opportunity to rethink federal land ownership and management as a catalyst for urban regeneration, civic stewardship, and housing production. Using the James V. Forrestal Building as a focal case study, the research examines the historical, policy, and spatial dynamics that have led to the current moment of reckoning. Located on Independence Avenue SW, straddling 10th Street between the National Mall and the Wharf, Forrestal is emblematic of the postwar federal design ethos: monumental, inward-facing, and hostile to street life. Once a symbol of bureaucratic permanence, the building now stands as a physical and symbolic barrier to urban connectivity and civic vitality. The case of Forrestal is used to explore broader questions: How can the federal government dispose of surplus property more effectively? What policy tools exist—or are needed—to unlock value and enable redevelopment? And what role should cities play in shaping the outcomes of federal land disposition? The thesis employs a mixed-methods approach that includes policy analysis, stakeholder interviews, precedent case studies, and spatial analysis of Southwest D.C. The work identifies a range of obstacles to effective disposition, including Title V of the McKinney-Vento Homeless Assistance Act, opaque OMB budget scoring rules, jurisdictional fragmentation, and the absence of a coordinating authority across federal agencies. It also identifies key lessons from successful projects such as The Yards, Walter Reed, and the Volpe Center, where thoughtful structuring and strong federal-local partnerships enabled transformative redevelopment of surplus land. The thesis concludes with ten detailed recommendations for reform, including reauthorization of the Federal Assets Sale and Transfer Act (FASTA), modernization of Title V and OMB scoring, the creation of Federal Redevelopment Zones, and the prioritization of housing, civic infrastructure, and design quality in disposition strategy. It argues that the federal government must shift from a passive landlord to an active steward of public land—one that collaborates with cities, integrates public benefit, and reflects democratic values through the built environment. In this moment of shifting federal needs, declining office demand, and urban transformation, the question is not whether federal real estate reform is needed—it is whether we will seize the opportunity. The fate of buildings like Forrestal will shape not only the skyline of Washington, D.C., but also the federal government’s legacy in America’s cities for generations to come.

Analyst Incentives

Mon, 01 Sep 2025 00:00:00 GMT

Analyst Incentives Green, Brice Analyst forecasts have been shown to reflect substantial behavioral biases and predict a number of macroeconomic phenomena. While we typically treat reported forecasts as statistical expectations, under uncertainty the reported point estimate will be sensitive to the payoff structure facing the forecaster. Using data on careers from LinkedIn, I describe the incentive structures faced by analysts, shedding light the extent to which pay and career success are tied to performance. Further, I extend a causal estimator to identify credible counterfactual forecasts and provide tentative causal evidence of the relationship between forecast errors and promotions.

Maverick Neuroscientist: What Does a Life in Science Look Like Outside the Ivory Tower? : Dr. Eugenio Vargas-Pena is a renowned psychiatrist in Paraguay, who conducts neuroscience research without university ties, funding, or peer review. ls his embodiment of the gentleman scientist an alternate path for those who want to break away from institutionalized science?

Mon, 01 Sep 2025 00:00:00 GMT

Maverick Neuroscientist: What Does a Life in Science Look Like Outside the Ivory Tower? : Dr. Eugenio Vargas-Pena is a renowned psychiatrist in Paraguay, who conducts neuroscience research without university ties, funding, or peer review. ls his embodiment of the gentleman scientist an alternate path for those who want to break away from institutionalized science? Chomik-Morales, Jessica This longterm narrative investigates the life and work of Dr. Eugenio Vargas-Peña, a neuropsychiatrist in Asunción, Paraguay who built a fully functional lab in his countryside home. Vargas-Peña conducts brain research independently, guided by decades of self-study, clinical practice, and an unwavering belief in the value of curiosity-driven inquiry. The piece interweaves historical context, character study, and personal narrative, using the author's own background in neuroscience and science communication to frame an inquiry into legitimacy, recognition, and alternative pathways in science. It asks: What defines a scientist today? Who gets to decide which ideas are taken seriously? And what are the consequences-creative or catastrophic-of working outside institutional boundaries? Through the lens of one man's eccentric yet earnest intellectual journey, this thesis invites broader reflection on the pressures shaping contemporary research and the enduring romance of unorthodox scholarship.

Machine Learning Systems for Unsupervised Time Series Anomaly Detection

Mon, 01 Sep 2025 00:00:00 GMT

Machine Learning Systems for Unsupervised Time Series Anomaly Detection Alnegheimish, Sarah Modern assets – from launched satellites to electric vehicles – output dense, multivariate time series data that must be monitored for deviations from “normal” behavior. This monitoring task is referred to as time series anomaly detection. The current state of the industry still depends on fixed or heuristic thresholds that often drown operators in false alarms, and can miss the subtle, context-dependent faults that matter most. This thesis addresses unsupervised time series anomaly detection as an end-to-end problem, asking how we can learn, evaluate, and deploy models that judiciously flag anomalies while remaining intuitive to the end user. This thesis provides contributions in the form of both algorithms and systems. First, it introduces three models that enlarge the design space of unsupervised time series anomaly detection: TadGAN, which leverages adversarial reconstruction; AER, which unifies predictive and reconstructive objectives in a single hybrid score; and MixedLSTM, which explicitly incorporates interdependencies to improve anomaly detection in multivariate time series. We propose two range-based evaluation metrics that quantify detection quality over temporal intervals. Second, it presents our system Orion, which abstracts anomaly detection pipelines as directed acyclic graphs of reusable primitives, providing user-friendly APIs and enabling interactive visual inspection. Building on this infrastructure, OrionBench performs periodic, fully reproducible benchmarks, producing leaderboards that align research innovations with the needs of end users. Third, the thesis explores a new paradigm – foundation models for unsupervised time series anomaly detection – by formulating SigLLM , which employs large language models and time series foundation models for zero-shot anomaly detection via prompting and forecasting. This paradigm indicates a promising path to developing scalable models for anomaly detection. Finally, beyond evaluating our systems on publicly available datasets, we provide extensive experiments on two industrial case studies that demonstrate improved detection accuracy and practical usability of our system.

Techniques for Reliability and Robustness in Integrated Electronic and Photonic Systems

Mon, 01 Sep 2025 00:00:00 GMT

Techniques for Reliability and Robustness in Integrated Electronic and Photonic Systems Chakraborty, Uttara Reliability and robustness are key concerns in the development of novel electronic and photonic materials, devices, and systems. This thesis presents statistical and machine learning techniques for reliability analysis of heterogeneously-integrated systems, extraction of variations from photonic test structure measurements, making smart decisions about test configurations in the face of time and resource constraints, and robust design of photonic components. To estimate reliability model parameters from lifetime datasets where multiple underlying failure mechanisms are present, a differential evolution framework and a boundconstrained expectation maximization algorithm are developed; both these approaches significantly outperform the gradient-based L-BFGS-B algorithm. New schemes for strategic failure analysis on a subset of the failed units are presented, both for detecting the presence of a second failure mechanism and for improving two-mechanism reliability models. A regression-based protocol is also presented for optimally selecting reliability test conditions to verify physical failure mechanism models. A maximum-likelihood-estimation-based approach is demonstrated for the simultaneous extraction of waveguide index and thickness variations using integrated photonic direction couplers and Mach-Zehnder interferometers. Schemes are proposed for optimal selection of cut-back test structures and for propagation loss estimation with a Bayesian prior distribution for fiber-coupling error. Finally, a robust Bayesian optimization algorithm using a new tunable acquisition function is presented for photonic component design. The methods developed in this thesis are expected to be broadly applicable to a wide variety of electronic and photonic devices and systems.

Synthesis and Applications of Large-Area Monolayer Graphene

Mon, 01 Sep 2025 00:00:00 GMT

Synthesis and Applications of Large-Area Monolayer Graphene Wang, Zhien (Abigail) Graphene, renowned for its exceptional electrical, mechanical, and chemical properties, is a promising candidate for next-generation electronics, photonics, and biosensing. However, realizing its full potential depends critically on the ability to synthesize high-quality monolayer graphene. In this thesis, we present a robust chemical vapor deposition (CVD) approach for synthesizing large-area, adlayer-free, single-orientation graphene on Cu(111) foil and Cu(111) film/sapphire. A comparative analysis between these two substrates reveals critical differences in wrinkle density, grain size, and strain — offering insights for optimizing graphene growth. We further identify and characterize defective merging behavior in single-orientation graphene domains. Contrary to conventional assumptions, these merging regions contain permeable defects, revealing previously unrecognized limitations in using single-orientation stitched graphene as an impermeable barrier. To scale up production while reducing human error, we also develop an autonomous CVD platform with automated sample handling, growth and post-growth oxidation. This system enables high-throughput and reproducible graphene synthesis with minimal supervision. Building on these synthesis advances, we explore multiple applications of large-area monolayer graphene. We discover that graphene can promote interfacial oxidation of metals like aluminum and titanium during deposition, whereas metals such as nickel remain stable — a finding that informs the engineering of metal-graphene interfaces for electronic devices. In parallel, we explored diverse applications of graphene, including its role as a transparent, flexible electrode in organic solar cells, along with several collaborative efforts demonstrating its use as a sensor for cardiac microtissues, and as a tunable microheater in mid-infrared devices. Altogether, this work advances both the fundamental understanding and technological scalability of monolayer graphene, positioning it as a versatile platform for future applications across electronics, optoelectronics, and biointerfaces.

Mapping signaling networks and rapidly evolving genes in the developing Arabidopsis seed at single-nucleus resolution

Mon, 01 Sep 2025 00:00:00 GMT

Mapping signaling networks and rapidly evolving genes in the developing Arabidopsis seed at single-nucleus resolution Martin, Caroline A. Seeds are an exceptional evolutionary innovation that enables the conditional allocation of maternal resources to successfully fertilized ovules. During early development, seeds accumulate nutrients that are utilized either by the embryo or by humans who harvest seed crops for food, biofuel, and livestock feed. Moreover, the grains of maize, rice, and wheat provide approximately 60% of the calories consumed worldwide. Although seeds are a cornerstone for ecosystems and modern agriculture, fundamental aspects of their development are incompletely understood. In this thesis, I develop a transcriptional atlas of seed development using the model plant Arabidopsis thaliana to clarify the functional compartmentalization, diversity, and developmental dynamics of cell types in the seed. I focus my analyses on how seed cell types communicate with one another to ensure successful propagation, and how genetic conflicts in the seed may drive rapid evolution in specific cell types. After characterizing the extent of short, secreted peptide expression in specific seed cell types, I perform in silico screens to match potential peptide hormones with their receptors. In total, I show that the seed coat shows functional compartmentalization around the gateway for maternal resources into seeds, that seed genes differentially expressed in a maternal resource transfer structure are rapidly evolving, and that genes underlying brassinosteroid biosynthesis and response are expressed in adjacent tissues, among other findings. This thesis illuminates potentially new mechanisms for inter-tissue coordination and provides a transcriptional reference for future seed studies.

Optimization in Deep Learning: Structured, Realistic and Interpretable Learning for Decision-Making

Mon, 01 Sep 2025 00:00:00 GMT

Optimization in Deep Learning: Structured, Realistic and Interpretable Learning for Decision-Making Tsiourvas, Asterios In recent years, deep learning has emerged as a powerful tool for data-driven decisionmaking. However, its adoption in high-stakes applications is often constrained by challenges related to interpretability, fairness, and generalization in structured or complex environments. This thesis develops new optimization methodologies to enhance the realism, structureawareness, and interpretability of deep learning models in decision-making tasks. We begin, in Chapter 2, by addressing the challenge of optimizing trained neural networks for data-driven decision-making. Although neural networks can encode rich representations of preferences or outcomes, directly optimizing their outputs can be computationally intractable and often may produce unrealistic prescriptions. We introduce scalable algorithms that leverage the piecewise-linear structure of ReLU networks, reducing the original hard-to-solve mixed-integer program into tractable linear programs. To ensure realism, we introduce constraints that restrict decisions to lie on the data manifold. We then extend this framework to any differentiable neural network or MIP-expressible model and show that it scales for networks with millions of parameters. In Chapter 3, we focus on decision-making under observational data. First, we study personalized treatment recommendations under discrete treatments. We introduce the Prescriptive ReLU (P-ReLU) network, a piecewise-linear model that partitions the input space into polyhedral regions, assigning treatments uniformly within each, and that can be translated into an equivalent interpretable decision tree. We demonstrate that P-ReLU achieves strong prescriptive accuracy and accommodates structural/prescriptive constraints with ease. Next, we consider the problem of large language model (LLM) routing, where a query must be dynamically routed to the best model under competing metrics like accuracy and cost. We develop a causal, end-to-end approach that learns routing policies directly from logged observational data, minimizing directly decision-making regret. Finally, we tackle the problem of generating realistic, manifold-aligned counterfactual explanations. To address this problem, we present a MIP formulation where we explicitly enforce manifold alignment by reformulating the highly nonlinear Local Outlier Factor (LOF) metric as a set of mixed-integer constraints. To address the computational challenge, we leverage the geometry of the network and propose an efficient decomposition scheme that reduces the initial hard-to-solve problem into a series of significantly smaller, easier-to-solve problems. We further extend this framework to any differentiable neural network or MIP-expressible machine learning model. In Chapter 4, we focus on structured machine learning. We first address the problem of hierarchical time series forecasting, where predictions must be both accurate and consistent with the aggregation structure of the hierarchy. While prior methods rely on fixed projection matrices, we propose learning the optimal oblique projection directly from data. The proposed end-to-end approach jointly trains the forecasting model and projection layer, significantly improving accuracy and coherence. Next, we study the problem of creating a highly expressive, interpretable, and fair machine learning model. We propose Neural-Informed Decision Trees (NIDTs), a model that combines the predictive power of neural networks with the inherent interpretability of decision trees. NIDTs use axis-aligned splits on dataset features to form transparent decision paths, and at each leaf, apply a linear predictor based on both the original features and neural embeddings from a task-specific network to capture non-linearities. To generate NIDTs, we develop a decomposition training scheme that supports direct integration of fairness constraints via a constrained convex optimization problem solved at each leaf. Finally, in Chapter 5, we address fairness and efficiency in emergency department (ED) operations, where prolonged length of stay (LOS) has been linked to adverse outcomes such as increased mortality and higher risk of hospital-acquired infections. We focus on the patient prioritization and placement aspects of ED operations to improve throughput and reduce wait times. We propose a novel MIP predictive-prescriptive framework that decomposes predicted LOS into actionable components, enabling a more granular and operationally meaningful model of ED dynamics. Fairness considerations are explicitly incorporated into the formulation. To address uncertainty, we introduce a sampling-based solution approach. Our method increases ED throughput by 50–100% and reduces average wait time by 50–75%, depending on current utilization levels, while achieving near-optimal performance compared to a clairvoyant oracle. This work was conducted in collaboration with a major U.S. academic medical center. To facilitate practical implementation, we also design an interpretable metamodel that approximates the predictive-prescriptive algorithm with high fidelity. Together, these contributions provide a unified perspective on deep learning for reliable decision-making, grounded in optimization and encompassing interpretability, structure-awareness, and causal reasoning, well-suited for high-stakes operational environments.

Reverberation Mapping of Supermassive Black Holes using Machine Learning

Mon, 01 Sep 2025 00:00:00 GMT

Reverberation Mapping of Supermassive Black Holes using Machine Learning Lewin, Collin Accreting supermassive black holes at the centers of galaxies, known as active galactic nuclei (AGN), offer a unique window into the physics of accretion and feedback that shape galactic evolution. Yet, the small spatial scales of these regions remain inaccessible to direct imaging. Reverberation mapping circumvents this limitation by using time delays between correlated emission at different wavelengths to infer physical size scales. While X-ray reverberation probes the innermost accretion flow, continuum reverberation in the UV, optical, and infrared (UVOIR) traces reprocessing by the accretion disk and broad-line region (BLR). In this thesis, I develop and apply frequency-domain timing techniques based on Gaussian Process (GP) regression to study AGN reverberation across X-ray and UVOIR regimes. By modeling the empirical variability of AGN light curves with GPs, I interpolate onto an evenly sampled time grid, enabling robust estimation of Fourier-resolved time lags despite irregular sampling or large time gaps. I apply this method to NuSTAR observations of the Narrow-line Seyfert 1 galaxy Ark 564, introducing a multi-task GP model that jointly learns kernel hyperparameters across light curves. This enables the first simultaneous modeling of lag and flux spectra from both NuSTAR and XMM-Newton using a relativistic reverberation model to constrain black hole mass and disk properties. Recent reverberation campaigns with the Neil Gehrels Swift Observatory and ground-based telescopes have revealed significant discrepancies between observed inter-band lags and standard accretion disk theory. These include unexpectedly large lag amplitudes (the “accretion disk size problem”) and weak correlations between X-ray and UV/optical light curves. To investigate further, I analyze recent Swift campaigns of Mrk 335 and Mrk 817 using GP-based frequency-resolved lag analysis. In both sources, standard disk lags appear only on short timescales (high frequencies), while longer-than-expected lags dominate at low frequencies. These lag excesses are consistent with reprocessing at larger radii, similar to the BLR. Mrk 817 offers a rare opportunity to connect the inner and outer accretion flow: I detect the first simultaneous measurement of X-ray and UVOIR lags, effectively mapping the full disk. These lags vary significantly over the campaign, with longer delays during periods of stronger X-ray obscuration. This suggests that a disk wind may modulate the observed lags by introducing additional reprocessing and/or blocking ionizing flux from reaching more-distant material. To test this obscuration effect across a population, I conduct the first statistical study of UV/optical lag excess versus physical parameters across the Swift campaigns. The results show that the lag excess is driven entirely by obscured AGN, while the lags of unobscured sources are, on average, consistent with thin-disk theory. Regression analysis reveals that X-ray column density explains over 80% of the variance in lag excess. As for the X-ray/UV connection, obscured AGN also tend to show weaker correlations and more variable lags, suggesting that line-of-sight absorption not only contributes additional reprocessed emission that extends the UV/optical lags, but may also decouple or delay the X-ray and UV variability. To make GP-based time series analysis accessible to the community, I developed the STELA Toolkit, a fully documented Python package for computing frequency-domain data products using GPs. I also benchmark GP performance against other interpolation methods, including state-of-the-art transformers, paving the way for scalable, ML-enabled timing analysis in the era of time-domain surveys like Vera Rubin.

Learning Nonlinear Dynamics: Methods and Applications

Mon, 01 Sep 2025 00:00:00 GMT

Learning Nonlinear Dynamics: Methods and Applications Rossi, Baptiste T. Accurate modeling of dynamical systems through differential equations is essential for scientific prediction and prescriptive control. Traditional model development, which relies on expert knowledge, parameter fitting and validation, is often iterative, time-consuming, and complicated by real-world data complexities such as noise and missing observations. This thesis addresses these challenges by developing robust, scalable, and interpretable methods for learning nonlinear ordinary differential equations (ODEs) and partial differential equations (PDEs) directly from data, with a particular emphasis on applications in fluid dynamics. In Chapter 2, we introduce a novel methodology for learning arbitrary nonlinear ODEs using collocation methods combined with interpolation. This approach demonstrates enhanced robustness to noise and significant computational speed-ups compared to classical system identification techniques, including the popular SINDy framework. It also provides a constructive method for reconstructing unobserved system components, making it applicable to partially observed systems, and offers theoretical guarantees on accuracy traditionally absent in strong-form identification. In Chapter 3, we combine the approach from Chapter 2 with sparse regression to derive sparse ODEs from data, demonstrating enhanced robustness to observational noise. Our method shows improved performance in recovering the true structures and coefficients on canonical benchmark tests under significant noise, while the performance of traditional surrogate methods deteriorates even with minimal noise. In Chapter 4, we extend this methodology to Partial Differential Equations (PDEs) using the method of lines, addressing issues related to data scale and interpolation ill-posedness. With a focus on Computational Fluid Dynamics (CFD), we show that our method goes beyond recovering complex nonlinear PDEs, such as the Navier-Stokes equations, from simulation data. The method can also be used as an a-posteriori indicator of simulation quality, providing insights into the effective PDEs represented by a given simulation, and pinpointing error-generating areas to inform adaptive mesh techniques. Lastly, in Chapter 5, we introduce a novel data-driven framework for modeling turbulent phenomena, a long-standing challenge in aerospace and climate science. Our approach addresses the Reynolds-Averaged Navier-Stokes (RANS) closure problem, which involves modeling the unobserved eddy viscosity field. We tackle two interconnected inverse problems: reconstructing the eddy viscosity from flow data and discovering its governing partial differential equations (PDEs), thereby proposing a new pathway to uncover new or refined RANS closure models directly from high-fidelity simulations. This chapter establishes a tractable baseline using a composite loss function, which we evaluate on canonical turbulent flows. Our results demonstrate that while the approach can recover governing equations when the ground truth eddy viscosity is known, significant challenges remain due to noise and numerical errors. We conclude that a more advanced reconstruction methodology is essential for robustly discovering these models, underscoring the potential of this data-driven approach and identifying critical areas for future research.

The Electronic Compressibility of Rhombohedral Graphene Multilayers

Mon, 01 Sep 2025 00:00:00 GMT

The Electronic Compressibility of Rhombohedral Graphene Multilayers Aronson, Samuel H. In condensed matter systems, energy bands with narrow dispersion frequently host correlated electronic phases that arise from strong Coulomb interactions. When these bands also have concentrated Berry curvature, the correlated phases may be topologically non-trivial. The low-energy bands of rhombohedral graphene multilayers possess both of these ingredients, making this a promising class of materials in which to search for correlated topological electronic ground states. This thesis describes our electronic compressibility measurements on rhombohedral graphene multilayers, with a particular focus on the pentalayer system (R5G). We utilize a planar capacitance technique that probes the thermodynamic density of states and enables us to extract energy gaps of incompressible phases. We observe a variety of correlated electronic phenomena including half and quarter metals, layer antiferromagnetism, correlation-driven Chern insulators, and thermodynamic signatures of potential Wigner crystallization. We also study the electronic compressibility of R5G aligned to a hexagonal boron nitride (hBN) substrate to form a moiré superlattice. Motivated by the recent discovery of the fractional quantum anomalous Hall effect in this system when the electrons are pushed away from the moiré interface by an external electric displacement field, we study the opposite moiré-proximal limit, in which the superlattice potential is considerably stronger. We observe integer and fractional Chern insulator states that persist down to low magnetic fields in addition to numerous trivial and topological charge density waves. We map out a phase diagram that is highly sensitive to both displacement and magnetic fields, establishing the R5G-hBN superlattice as a highly-tunable system for studying the interplay between intrinsic band topology and strong lattice effects.

Hadronic Structure with Classical and Quantum Computing

Mon, 01 Sep 2025 00:00:00 GMT

Hadronic Structure with Classical and Quantum Computing Avkhadiev, Artur Calculations in lattice quantum chromodynamics (QCD) — presently the only known systematically improvable approach to describe the strong nuclear force in the nonperturbative regime from first principles — are playing an increasingly important role in revealing how hadrons emerge from the interactions of the underlying degrees of freedom: quarks and gluons. With computational and theoretical advances, more fruitful connections have emerged between lattice QCD and phenomenology, and the field is now well into a stage ripe for deriving tighter constraints on hadronic structure through joint analyses of numerical lattice QCD results with experimental data. This thesis summarizes lattice QCD calculations of the Collins-Soper (CS) kernel: a nonperturbative function whose inclusion in joint analyses has the potential to advance the study of multidimensional hadronic structure. The CS kernel is an anomalous dimension of transverse-momentum-dependent (TMD) distributions describing a three-dimensional structure of ultrarelativistic hadrons as a function of quark-gluon momenta collinear with and transverse to the hadron's motion. Constraints on the CS kernel at nonperturbative transverse-momentum scales are instrumental to relate TMDs across scales and processes. The kernel differs for quark and gluon TMDs, but is otherwise universal. This thesis presents the first lattice QCD determination of the quark CS kernel with systematic control over operator mixing, quark mass, and lattice discretization, and a proof-of-principle lattice calculation of the gluon CS kernel providing the first nonperturbative constraints on this quantity. Additionally, this thesis summarizes exploratory studies on how Hamiltonian calculations — realized with quantum-computer simulations and tensor networks — may be combined with conventional Monte Carlo calculations based on Lagrangian formulations in Euclidean space. These studies examine how constructions of interpolating operators, used in conventional calculations to map between the vacuum and a ground state of interest, may be optimized in Hamiltonian calculations to increase overlap with the target state. Results, limited to the Schwinger model, support further investigations of this approach in theories more closely resembling QCD as quantum-computing and tensor-network technologies continue to mature.

Accelerating RTL Simulation Through Fine-grained Task Dataflow and Selective Execution

Thu, 01 Jun 2023 00:00:00 GMT

Accelerating RTL Simulation Through Fine-grained Task Dataflow and Selective Execution Elsabbagh, Fares Fast simulation of digital circuits is crucial to build modern chips. Current processors and SoCs integrate hundreds of complex components, including cores, accelerators, and memory hierarchies. Simulating these systems is necessary to verify correctness and explore the design space. Simulation can happen at different levels of abstraction. In this work we focus on Register-Transfer-Level (RTL) simulation. While RTL simulators are frequently used in development due to their quick compilation times, their runtime performance is slow. This is because as the designs are scaled up, multicore communication and scheduling overheads limit performance and scalability. We present ASH, a parallel architecture tailored to RTL simulation workloads. ASH consists of a tightly codesigned hardware architecture and compiler for RTL simulation. ASH exploits two key opportunities. First, it performs dataflow execution of small tasks to leverage the fine-grained parallelism in simulation workloads. Second, it performs selective event-driven execution to run only the fraction of the design exercised each cycle, skipping ineffectual tasks. ASH hardware provides a novel combination of dataflow and speculative execution, and ASH’s compiler features several novel techniques to automatically leverage this hardware. We evaluate ASH in simulation using large Verilog designs that represent different types of architectures. With 256 simple cores, ASH is gmean 1,485× faster than 1-core Verilator, and it is 32× faster than Verilator on a server CPU with 32 complex cores while using 3× less area.

Task Scheduling Techniques to Accelerate RTL Simulation

Thu, 01 Jun 2023 00:00:00 GMT

Task Scheduling Techniques to Accelerate RTL Simulation Sheikhha, Shabnam Fast simulation of digital circuits is crucial to build modern chips. Slow simulation lengthens chip design time and makes bugs more frequent. While simulation can happen at different levels of abstraction, Register-Transfer-Level (RTL) simulation is the usual bottleneck in chip design, as it is needed for ongoing debugging and evaluation. Current simulators scale poorly across CPU cores, because they are unable to exploit the fine-grained parallelism inherent in simulation workloads. We present ASH, a parallel architecture tailored to simulation workloads. ASH consists of a tightly codesigned hardware architecture and compiler for RTL simulation. ASH exploits two key opportunities. First, it performs dataflow execution of small tasks to leverage the fine-grained parallelism in simulation workloads. Dataflow execution exposes abundant parallelism, as each task can run as soon as its inputs are available. Second, it performs selective event-driven execution to run only the fraction of the design exercised each cycle, skipping ineffectual tasks. Selective execution introduces dynamic data dependences since skipped tasks do not communicate data. ASH employs speculative execution to handle these dependencies. ASH’s hardware provides a novel combination of dataflow and speculative execution, and ASH’s compiler features several novel techniques to automatically leverage this hardware. The key compiler techniques include a novel partitioning for minimizing data communication while maintaining load balance, and a strategic coarsening mechanism to reduce the overheads of fine-grained tasks. We evaluate ASH in simulation using large Verilog designs. With 256 simple cores, ASH is gmean 1,485× faster than 1-core Verilator, and it is 32× faster than Verilator on a server CPU with 32 complex cores while using 3× less area.

Scheduling Strategies for Bus Operator Retention: A Mixed-Methods Evaluation of Bus Operator Preferences and 4-Day Workweek Feasibility

Mon, 01 Sep 2025 00:00:00 GMT

Scheduling Strategies for Bus Operator Retention: A Mixed-Methods Evaluation of Bus Operator Preferences and 4-Day Workweek Feasibility Baum, Amelia Rose Public transit agencies face significant and growing challenges related to workforce shortages, absenteeism, and employee retention, which threaten service reliability. Reports found that 90% of U.S. transit agencies are experiencing a workforce shortage, with 84% claiming that the shortage affects their ability to provide scheduled service. Industry-wide, operator absence is a significant contributor to missed work at transit agencies nationwide and has, in many cases, delayed the full reinstatement of service at transit agencies following the COVID-19 pandemic. The quality of bus operators' work is significantly impacted by inflexible crew scheduling constraints. However, most studies focus on pay, benefits, and infrastructure, neglecting the importance of scheduling. This thesis aims to fill this gap by examining the potential for crew scheduling improvements to enhance the quality of life for bus operators through a three-part case study at the Chicago Transit Authority. Part 1 analyzes the historical work preferences of CTA bus operators, providing actionable insights for scheduling improvements. Part 2 presents a high-fidelity proof of concept in HASTUS, using block schedules (10-hour-a-day runs that are intended to be run by an operator 4 days a week) and rostering to reduce negative work traits, increase consecutive and weekend days off for most operators, while maintaining schedules for the top 20% of senior operators. Part 3 evaluates the new 10-hour, 4-day-per-week packaged schedules via an LLM-based paired alternatives survey of operators at one CTA garage, measuring the desirability of the proof of concept and collecting qualitative feedback. Overall, the new schedules substantially improve the quality of work for operators by guaranteeing at least one weekend day off, at least two consecutive days off, and increasing day-to-day schedule consistency and overnight rest time, while maintaining constant vehicle requirements and total pay hours. The survey results show that 72% of operators at the 74th Street garage support the new schedule paradigm, demonstrating strong support for their potential adoption and encouraging future exploration of a block schedule hybrid rostering paradigm at the CTA and other transit agencies.

A Charcuterie Platter of QCD Matter

Mon, 01 Sep 2025 00:00:00 GMT

A Charcuterie Platter of QCD Matter Sun, Zhiquan One of the greatest current challenges in theoretical high energy physics is to understand the dynamics of Quantum Chromodynamics (QCD). In this thesis, I address a variety of questions in QCD using Effective Field Theory (EFT). The first question deals directly with the observed phenomenology of QCD: How can we use EFT to disentangle the complicated three-dimensional dynamics of how quarks and gluons, the fundamental degrees of freedom of QCD, combine to form the observed bound states in nature called hadrons? I initiate a new formalism using Heavy Quark Effective Theory to study this dynamical process known as hadronization. I shed new light on the transverse momentum-dependent fragmentation process of heavy (charm and bottom) quarks by making use of the fact that heavy quarks with masses much larger than the strong interaction scale decouple from the rest of the hadronization cascade. I also present exciting heavy quark phenomenology at existing colliders and the upcoming Electron-Ion Collider. The second question investigates the field theory structure of QCD: What can we learn about the nonperturbative structure of the quantum field theory through the abstruse emergent phenomenon in QCD called “confinement”, which traps quarks and gluons inside hadrons? I study a class of cleverly constructed observables known as energy correlators by using fieldtheory based methods to determine the leading nonperturbative contribution, and examine the universality of the nonperturbative matrix element that gives rise to this contribution. I also show that including the nonperturbative contribution has a significant impact on the extraction of the strong coupling constant, a fundamental parameter of the Standard Model, using tools such as factorization and resummation from EFT. Last but not least, the final question explores the underlying symmetry properties of QCD and its potential completions: How robust is the axion solution to the strong CP (ChargeParity) problem, and what are some of its implications beyond the realm of QCD? I examine the axion quality problem in post-inflationary QCD axion models with different symmetry properties and identify a new tension with standard cosmology. I further show that the axion string-domain wall dynamics is more complicated than commonly expected, undermining the reliability of a unique mass prediction for axion dark matter in post-inflationary models. I showcase the importance of considering both high-energy extensions and the EFT at low energy, and uncover new complexity of the axion solution to the strong CP problem.

Learning from pre-pandemic data to design and test future-proof therapeutics

Mon, 01 Sep 2025 00:00:00 GMT

Learning from pre-pandemic data to design and test future-proof therapeutics Gurev, Sarah Effective pandemic preparedness relies on predicting immune-evasive viral mutations to enable early detection of variants of concern and design vaccines and therapeutics that are resilient to future viral evolution. However, current strategies for viral evolution prediction are not available early in a pandemic and have limited predictive power – experimental approaches require host polyclonal antibodies and existing computational methods draw heavily from current strain prevalence. In addition, vaccines and therapeutics have been designed with an eye towards past or circulating variants, not towards future evolution. To address these challenges, we developed EVEscape, a generalizable framework that integrates fitness predictions from a deep generative model of evolutionary sequences with biophysical and structural information. EVEscape quantifies the immune escape potential of viral strains at scale and is applicable before surveillance sequencing, experimental scans, or 3D structures of antibody complexes are available. We demonstrate that EVEscape, trained on sequences available prior to 2020, performs as accurately as high-throughput experimental scans at anticipating pandemic variation for SARS-CoV-2 and is generalizable to other viruses including Influenza A virus, HIV, and understudied viruses with pandemic potential such as Lassa and Nipah. We investigate both alignment-based and protein language models to explore the best model of mutation effects across pandemic-threat viral families. We demonstrate the utility of EVEscape in three critical applications: (1) Surveillance efforts flagging high escape SARS-CoV-2 variants from their first appearance (2) Design of panels of viral antigens that mimic future viral variants for early, proactive evaluation of the future protection of vaccines and therapeutic; and (3) Design of a pan-sarbecovirus nanoparticle-based vaccine capable of eliciting broad, long-lasting protection against sarbecoviruses, including future variants. This three-pronged approach represents a paradigm shift in pandemic preparedness, offering a novel strategy to preemptively address viral families with pandemic potential and significantly bolster global prevention efforts.

Topics in quantum information theory and quantum many-body physics

Mon, 01 Sep 2025 00:00:00 GMT

Topics in quantum information theory and quantum many-body physics Balasubramanian, Shankar In this thesis we present two results relating to the intersection between quantum information theory and quantum many-body physics. The first pertains to quantum algorithms, where few computational problems are believed to exhibit exponential separation between quantum and classical performance. For those that are, natural generalizations remain elusive. One speedup that has especially resisted generalization is the use of quantum walks to traverse the welded tree graph, due to Childs, Cleve, Deotto, Farhi, Gutmann, and Spielman. We show how to generalize this to a large class of hierarchical graphs in which the vertices are grouped into “supervertices” that are arranged according to a d-dimensional lattice. Supervertices can have different sizes, and edges between supervertices correspond to random connections between their constituent vertices. The traversal time of quantum walks on these graphs are related to (a) the existence of small subspaces within which the quantum walk evolves and (b) the localization properties of the quantum walk within these subspaces. We find examples of hierarchical graphs that yield provable speedups over classical algorithms ranging from superpolynomial to exponential, depending on the underlying dimension and the random graph model. We also discuss how to relax criterion (a) to the existence of a small and approximate subspace by using techniques from graph sparsification. The second result pertains to fault-tolerant quantum memories. Storing a qubit in a noisy environment is crucial for developing full-scale quantum computers. While constructions of fault-tolerant quantum memories exist, they often assume that quantum operations are not local and assisting classical computation operates instantaneously and noislessly. In particular, constructing a topological quantum memory below four dimensions with local quantum and classical operations that is fault-tolerant under both quantum and classical noise is an open problem. We construct a local quantum memory for the 2D toric code using ideas from the classical cellular automata of Tsirelson and Gács. Our memory preserves a logical state for exponential time in the presence of both classical and quantum noise below a constant noise rate. While our 2D quantum memory is built from operations that depend on space and time, we construct a fault-tolerant quantum memory in 3D using stacks of 2D toric codes that can be built with time-independent operations.

Quantum Matter in the Era of Generalized Symmetries

Mon, 01 Sep 2025 00:00:00 GMT

Quantum Matter in the Era of Generalized Symmetries Chatterjee, Arkya The discovery of generalized symmetries has led to powerful new insights into quantum matter. They have been used to classify new families of quantum phases, place constraints on phases realizable in a given physical system, and conceptually unify seemingly disparate phenomena. In many ways, they prove just as powerful as traditional symmetries at organizing and constraining the theories that describe quantum matter. In this thesis, we attempt a unification of such constraints by developing a holographic correspondence between (generalized) symmetries and topological orders, called the Sym/TO correspondence. For any (finite internal) symmetry of a quantum system in d (spatial) dimensions, we associate with it a unique topological order in d + 1 dimensions, called its Symmetry Topological Order (SymTO). We devise an operator algebraic recipe to compute the SymTO data for any lattice spin model, demonstrating it in a number of examples. We then use the SymTO to classify possible quantum phases allowed by the symmetry—we call this a generalized Landau paradigm. Besides classifying phases, we also identify constraints on the phase transitions between them using a SymTO-resolved modular bootstrap. We test this framework in a quantum spin chain with non-invertible symmetries. We discover a new Kramers-Wannier-like duality and a rich phase diagram including a noninvertible symmetry-enriched incommensurate phase. The translation symmetry of the spin chain has a nontrivial interplay with the lattice Kramers-Wannier duality, which matches the anomaly of the corresponding non-invertible symmetry in the low-energy effective field theory. Finally, we explore such unusual anomaly-matching mechanisms in more detail in the context of the chiral anomaly of a single massless Dirac fermion, demonstrating a novel lattice realization of chiral symmetries and their anomaly.

Systems Materials Design of Ordered Nanocomposite Assemblies

Mon, 01 Sep 2025 00:00:00 GMT

Systems Materials Design of Ordered Nanocomposite Assemblies Thrasher, Carl James The ability to precisely organize matter across multiple length scales is a central challenge in modern materials science. In this dissertation, I develop a systems materials design approach to engineer hierarchically structured nanocomposite assemblies, integrating molecular recognition, supramolecular chemistry, colloidal assembly, and bulk processing into unified material platforms. At the molecular and nanoscale, I investigate how multivalent supramolecular interactions can be rationally programmed by controlling the architecture of polymer binders grafted to nanoparticle surfaces. Through systematic variations in polymer topology, recognition group density, and scaffold geometry, I demonstrate how polymer design dictates the thermodynamic strength and multivalency of nanoparticle superlattice assembly, enabling precise control of thermal stability, crystallographic symmetry, and collective bonding behaviors in massively multivalent systems. Building on these design principles, I develop a colloidal metallurgy framework to process selfassembled nanoparticle superlattices into dense macroscopic polycrystalline solids while preserving nanoscale order. By systematically studying the interplay of temperature, pressure, and time during colloidal sintering, I elucidate mechanisms of densification, defect evolution, and grain growth unique to colloidal systems, establishing processing–structure relationships that parallel but fundamentally diverge from atomic sintering. Finally, I extend these concepts to create stretchable nanocomposite supercrystals, embedding supramolecularly assembled superlattices into elastomeric matrices via co-engineered polymer chemistries that enable hierarchical strain transduction. These materials combine the nanoscale precision of supercrystals with mechanical robustness, reconfigurability, and stimuli-responsive optical properties, illustrating a scalable pathway to multifunctional metamaterials. Collectively, this work demonstrates how a systemslevel integration of molecular design, colloidal assembly, and bulk processing enables new paradigms for the synthesis of hierarchically ordered, functional nanocomposites.

The novel roles of BCL6 and BATF3 in regulating human CD8⁺ T cell dysfunction

Mon, 01 Sep 2025 00:00:00 GMT

The novel roles of BCL6 and BATF3 in regulating human CD8⁺ T cell dysfunction Traunbauer, Anna Katharina Reduced effector function and elevated inhibitory receptor expression are hallmarks of exhausted CD8⁺ T cells, yet the underlying molecular and epigenetic drivers remain incompletely defined. Here, we developed an in vitro repeated stimulation model to recapitulate features of human CD8⁺ T cell dysfunction and delineate transcriptional and epigenetic landscapes. Our analyses revealed that BCL6 and BATF3 are robustly upregulated in dysfunctional CD8⁺ T cells, with ATAC-seq demonstrating enhanced chromatin accessibility at their gene loci. Transcription factor footprinting shows increased BATF3 motif occupancy in chronically stimulated cells and integrative multi-omic analysis combining footprints, open chromatin regions, RNA-seq and ChIP-seq data revealed that putative BATF3 target genes may include master regulators of exhaustion. Moreover, overexpression of BCL6 or BATF3 markedly upregulates TIM-3 expression and suppressed cytokine release, establishing their capacity to induce T cell dysfunction. We further validated these findings ex vivo in antigen-specific CD8⁺ T cells from patients with advanced melanoma, as well as HCV and HIV infections, where cells were enriched for BCL6^high and BATF3^high subsets co-expressing canonical exhaustion markers such as PD-1, TIM-3 and CD39. Notably, Single-cell RNA sequencing of HIV-specific CD8⁺ T cells identified a distinct BCL6^high PD1⁻ progenitor population that gives rise to two distinct subsets via divergent differentiation trajectories: one branch generates effector-like BCL6^high PD1⁺ cells, whereas the other produces BCL6^high PD1⁺ cells that retain an exhaustion gene signature alongside partial memory-like feature. Collectively, these findings identify BCL6 and BATF3 as key mediators of human CD8⁺ T cell dysfunction and illuminate novel transcriptional and epigenetic pathways that may be leveraged for therapeutic intervention in cancer and chronic viral infections.

Aspects of Nonperturbative Heavy Quark Physics

Mon, 01 Sep 2025 00:00:00 GMT

Aspects of Nonperturbative Heavy Quark Physics Lin, Joshua The properties of charm and bottom quarks are an interesting corner of Quantum Chromo-Dynamics (QCD) due to the fact that their masses are much heavier than the typical QCD interaction energy ΛQCD. Due to this scale separation, it is possible to describe these heavy quarks by Effective Field Theories (EFTs) that simplify their equations of motion, make explicit additional symmetries that only appear for heavier quark masses, and simplify the theoretical calculations required for predictions. By discretising these EFTs in a lattice regularisation, nonperturbative calculations of observables of interest become possible. This thesis presents progress towards systematically controlled calculations of two such observables: the Spectator Effect contributions to the inclusive decay rates of b-hadrons, and the real-time dynamics of fermions propagating in a thermal medium. Standard EFT calculations in Lattice-QCD proceed by expressing observables as sums over perturbatively computed Wilson coefficients and nonperturbative matrix elements that can be calculated by path integral monte-carlo methods. Though it is possible to carry out this procedure within a regulator-independent renormalization scheme, in practice almost all such decompositions are computed in the modified minimal subtraction scheme MS which is only defined for the dimensional regulator (DR), due to its simplicity. Computing such observables therefore requires a matching between lattice regularised operators and operators renormalized in MS. In Chapter 2, both the dimensional regulator (DR) and the lattice regulator are reviewed, with a particular emphasis on techniques needed for calculations carried out in later sections. An interesting subtelty in DR is the need to introduce d-dimensional counterparts to the Dirac γ-matrices, which a-priori are only well defined in integer number of dimensions. This analytic continuation is of practical importance as it introduces additional Evanescent Operators (Sec. 2.1.4) that have physical consequences. In Sec. 2.1.5, traces of d-dimensional γ-matrices were related to Tutte polynomial evaluations [4], presenting a new graph-theoretic interpretation of the dimensionally regulated γ-matrices. One strategy of renormalizing lattice-regulated operators into MS involves first renormalizing into a regulator independent scheme, before perturbatively matching between the regulator independent scheme and MS. In Chapter 3, regulator independent position-space (X-space) schemes for renormalizing operators defined in the leading order Heavy Quark Effective Theory (HQET) are proposed [3]. Compared to other regulator independent renormalization schemes such as RI-xMOM, X-space schemes have the benefit that they are gauge invariant. The next to leading order matching calculations between X-space and MS are presented for heavy-light and heavy-light-light multiplicatively renormalizable operators, as well as ∆Q = 0 and ∆Q = 2 four quark operators relevant for heavy hadron decays and mixing, where Q refers to the static quark number. Due to their heavy masses, hadrons containing heavy quarks decay via the weak nuclear force. Experimental measurements of these lifetimes provide precision determinations of the fundamental parameters of the Standard Model. The Heavy Quark Expansion expresses the inclusive lifetimes of heavy hadrons in terms of matrix elements of HQET operators of increasing dimension. The Spectator Effects are contributions due to the ∆Q = 0 four-quark operators, where the light quark degrees of freedom within a heavy hadron participate in the decay. In Chapter 4, a Lattice-QCD determination of the static decay constant f HQET B and the isospin-nonsinglet portion of the Spectator Effect matrix elements for heavy-light mesons is presented. Fits of bare matrix elements were performed for three different lattice spacings, and renormalized with the schemes proposed in Chapter 3 before a continuum limit is taken. Due to the heavy masses mQ of the heavy quarks, it is possible to find temperatures T approximately satisfying a hierarchy ΛQCD ≪ T ≪ mQ. At these temperatures, QCD undergoes a deconfinement transition into the Quark-Gluon-Plasma (QGP) phase where the light degrees of freedom are no longer confined, and instead screen the long-range colour forces. The heavy quarks however are not thermalised, and act as probes of the QGP. Further understanding of the QGP requires first principles simulations of the heavy quark dynamics at finite temperature, however such calculations are difficult due to the enormous size of the Hilbert space. Variational approximations of the Hilbert space encode wavefunctions within a few parameters, and provide a practical method to simulate many particle systems. As a testcase, the variational approach was applied for the first time to simulate fermions at finite temperature in a simple QFT: the 1+1d U(1) gauge theory known as the massive Schwinger model. Both the real-time dynamics of string like states, and the properties of the thermal state were studied, and such variational methods are shown to be promising approaches to the more difficult case of a heavy quark effective theory in QCD.

Probing the Nonperturbative Physics of QCD with Normalizing Flows and a moderate number of Pions

Mon, 01 Sep 2025 00:00:00 GMT

Probing the Nonperturbative Physics of QCD with Normalizing Flows and a moderate number of Pions Abbott, Ryan William Quantum Chromodynamics (QCD) is a cornerstone of the standard model of particle physics, and the best known theory of strong nuclear interactions. The only known systematically improvable ab-initio method for accessing the nonperturbative physics of QCD is Lattice QCD is, and this thesis presents two advances in our understanding QCD using lattice-based methods. The first is a calculation using many-pion systems to map out the entire zero temperature, nonzero isospin density region of the QCD phase diagram. The calculation uses novel methods for working with many-pion systems that enables working with thousands of pions, and furthermore provides rigorous constraints on the baryon-dense region of the QCD phase diagram. The second is an application of methods from machine learning (namely normalizing flows) in order to accelerate sampling. This approach has the promise of eliminating issues such as critical slowing down, as well as introducing novel tools and methods that enable methods of calculation that would be possible otherwise.

Limits of QCD

Mon, 01 Sep 2025 00:00:00 GMT

Limits of QCD Gao, Anjie This thesis explores the fundamental kinematic limits of Quantum Chromodynamics (QCD), including the soft, collinear, and Regge limits, using soft-collinear effective theory (SCET). We begin by studying transverse momentum dependent (TMD) physics in semi-inclusive deep inelastic scattering (SIDIS), which probes the small transverse momentum regime arising from the soft and collinear limits of QCD. We derive all-order factorization theorems for azimuthal asymmetries in SIDIS at next-to-leading power (NLP). We also propose a new angular observable, q_∗, for probing TMD dynamics at the future Electron-Ion Collider (EIC), which enables an order-of-magnitude improvement in experimental resolution while retaining sensitivity to TMD distributions. Next, we apply the TMD formalism to a class of observables known as energy correlators. We study the transverse energy-energy correlator (TEEC) in the back-to-back limit, a dijet observable at hadron colliders, and the three-point energy correlator (EEEC) in the coplanar limit, a trijet observable at lepton colliders. For both observables, we derive allorder factorization theorems and resum large logarithms to next-to-next-to-next-to-leading logarithmic (N3LL) accuracy. Finally, we analyze the Regge limit of 2 → 2 QCD amplitudes. By factorizing these amplitudes into collinear jet and soft functions and studying their rapidity evolution, we define Regge-like anomalous dimensions in a gauge-invariant manner. At the level of the exchange of two Glauber gluons in the t-channel, we recover the BFKL equation from a purely collinear perspective. Extending to three-Glauber exchange, we derive the first closed-form renormalization group equations for Regge cut contributions in several nontrivial t-channel color representations, providing a systematic method for organizing non-planar QCD amplitudes at high energy.

Determining the Molecular Underpinnings of Iron Homeostasis in Human Cells

Mon, 01 Sep 2025 00:00:00 GMT

Determining the Molecular Underpinnings of Iron Homeostasis in Human Cells Lee, April Precise regulation of nutrient availability is crucial for cellular function and survival. Iron, in particular, is tightly regulated as it serves as an essential cofactor for numerous enzymes but can catalyze the formation of toxic radicals at elevated levels. To maintain the necessary cytoplasmic iron concentration, cells store excess iron in large proteinaceous cages called ferritin and, when available iron levels fall, they degrade these cages, liberating the stored iron for use. This thesis focuses on the molecular mechanisms underlying cellular iron sensing, as well as the molecular interactions supporting regulated ferritin degradation and subsequent iron release. Specifically, this work interrogates the protein interactions involved in ferritinophagy, a form of selective autophagy that leads to the lysosomal degradation of ferritin. Extending prior work that identified key components supporting ferritinophagy, including the selective autophagy receptor protein NCOA4 and its cognate autophagosomal receptor GATE16, experiments described here uncover the molecular contacts between these proteins. I found that NCOA4 bears two short linear motifs that each bind to GATE16 with weak affinity. However, these binding motifs are highly avid and, in concert, support high-affinity binding of NCOA4 to oligomerized GATE16. I further describe that ferritin degradation in cultured human cells relies on the contacts I identified biochemically. Moreover, I found that iron decreases NCOA4’s affinity for GATE16, providing a plausible mechanism for irondependent regulation of ferritinophagy. Taken together, this work suggests a general mechanism by which selective autophagy receptors can distinguish between inactive monomeric GATE16 and the active oligomerized forms that primarily drive autophagy. In related studies, I have biochemically probed the NCOA4•ferritin interface, with these experiments suggesting a novel function of NCOA4 in modulating ferritin cage structure – either through cage dismantling or through the formation of higher order structures. Taken together, these studies further define the molecular mechanisms by which NCOA4 aids cells in maintaining iron homeostasis, and they provide the requisite reagents for future work aimed at building a unified model for how mammalian cells regulate this vital but toxic metal.

Sampling Methods for Fast and Versatile GNN Training

Thu, 01 Sep 2022 00:00:00 GMT

Sampling Methods for Fast and Versatile GNN Training Alkhatib, Obada Graph neural networks (GNNs) have become a commonly used class of machine learning models that achieve state-of-the-art performance in various applications. A prevalent and effective approach for applying GNNs on large datasets involves mini-batch training with sampled neighborhoods. Numerous sampling algorithms have emerged, some tailored for specific GNN applications. In this thesis, I explore ways to improve the efficiency and expressivity of existing and emerging sampling schemes. First, I explore system solutions to facilitate the development of fast implementations of different sampling methods. I introduce FlexSample, a system for efficiently incorporating custom sampling algorithms into GNN training. FlexSample leverages the types of performance optimizations found in SALIENT, a state-of-the-art system for fast training of GNNs with node-wise sampling. In experiments with 4 GNN models which use layer-wise and subgraph sampling, FlexSample achieves up to 1.3× speed-up for end-to-end training over PyTorch Geometric with the same sampling code. Furthermore, FlexSample extends SALIENT with highly-optimized C++ implementations of FastGCN and LADIES layer-wise sampling, which achieve 2×–5× speed-up over their respective Python implementations. Second, I introduce a novel framework for learning neighbor sampling distributions as part of GNN training. Key components of this framework, which I name PertinenceSample, are: (i) a differentiable approximation of node-wise sampling for GNNs; and (ii) a parametrization of node sampling distributions as node- or edge-wise weights of attention-like GNN layers. I present an initial exploration of the potential of PertinenceSample for improving node classification accuracy in the presence of noisy edges. Specifically, in two synthetic experiments where roughly half of a node’s neighbors may have similar features but different labels, I demonstrate that extending a GraphSAGE model with a 2-layer perceptron for learning the PertinenceSample weights can improve classification accuracy from 50%–75% to (nearly) 100%.

Designing Electrocatalysts for the Production and Oxidation of Liquid Fuels

Mon, 01 Sep 2025 00:00:00 GMT

Designing Electrocatalysts for the Production and Oxidation of Liquid Fuels Zheng, Daniel J. With the ever-rising CO₂ levels in the atmosphere, it is paramount to cease reliance on fossil fuels to meet global energy demands. While the cost of electricity from renewable sources, such as solar and wind, continues to decrease and has even fallen below that of fossil fuels since 2014, these renewable energy sources suffer from intermittency, potentially causing shortages at peak demands. Thus, methods to store or economically use excess renewable energy are needed for full decarbonization. One promising avenue is to store the excess generated electrical energy in chemical bonds, creating molecules and materials with industrial or energy storage utility. In this proposed scheme, the renewable electricity would be used to electrochemically convert earth-abundant molecules into value-added chemical or fuels. These generated products could then be utilized as feedstocks in industrial applications or as a fuel source to generate electricity when needed by transforming back into their earth-abundant forms. Central to transforming earth-abundant molecules into value-added chemicals or fuels is the oxygen evolution reaction (OER), which is found in nearly every process. The plentiful nature of OER’s main reactant, water, and moderate thermodynamic potential of 1.23 V vs. the reversible hydrogen electrode, make OER an ideal reaction to pair with other transformations. However, the slow kinetics of OER significant hinder the efficiency of these processes. As such, discovering new OER catalysts with high activity and stability would have wide-spread impacts. On the other hand, one of the most promising renewable fuel sources is methanol, which boasts about 3 times the energy density of hydrogen and can be used as an alternative to hydrogen in proton exchange membrane fuel cells. However, the sluggish kinetics of the methanol oxidation reaction (MOR), even with current state-of-the-art noble metal catalysts causes direct methanol fuel cells to reach an efficiency of <40%, limiting their practical usage. While significant research has been invested in discovering new MOR electrocatalysts, PtRu has reigned for 5 decades, highlighting the need for a true breakthrough. In this thesis, electrocatalysts for OER and MOR are examined in depth. For OER, metal-hydroxide organic frameworks (MHOFs), a promising new class of hybrid organic-inorganic materials with potential to mimic the superior functionality of enzymes, are studied. Operando vibrational and absorption spectroscopy methods are used to characterize the degradation mechanisms and lattice oxygen exchange capacity as a function of the linkers. Using such knowledge, defects are engineered into the MHOF that increase both the activity and stability compared to the pristine material. Furthermore, the traditionally reported MOR mechanism is studied using isotope-labeled reactants and operando mass spectrometry. These experiments revealed that, in contradiction to typically accepted mechanisms, the C-O bond in methanol can be cleaved during MOR, with the resulting CO₂ molecule containing two water-derived oxygen atoms, opening a new paradigm for MOR catalyst design. Driven by the need to discover new materials at scale, a fluorescence-based OER catalyst screening method is developed that can screen an entire composition space simultaneously. In addition, an AI-driven, automated platform for screening a high-dimensional multimetallic space for MOR is presented.

Bridging the Gap: From Artificial Intelligence and Optimization Theory to Action

Mon, 01 Sep 2025 00:00:00 GMT

Bridging the Gap: From Artificial Intelligence and Optimization Theory to Action Petridis, Periklis S. Despite significant theoretical advances in Operations Research (OR) and Artificial Intelligence (AI), a persistent gap remains between these developments and their practical implementation in real-world settings. Despite significant progress in these fields, many OR and ML approaches struggle to scale to realistic problem sizes, lack robustness to uncertainty, or fail to address implementation constraints faced by practitioners in industry. Through four distinct works conducted in collaboration with industry partners, this research demonstrates how methodological advancements can bridge this theory-practice divide while maintaining rigorous theoretical foundations and guarantees. In the first part, we focus on optimization methodologies that scale traditional OR approaches to handle real-world problem sizes and uncertainty. In Chapter 2, we develop a stochastic Benders decomposition scheme for large-scale network design problems, a class of problems ubiquitous in logistics, transportation, and energy sectors. By incorporating sampling techniques within the decomposition framework, we achieve deterministic optimality guarantees while reducing computational costs, enabling solutions for networks with 700 nodes—an order of magnitude larger than previously tractable instances—while achieving optimality gaps of 5-7% compared to 16-27% for traditional deterministic approaches. In Chapter 3, we present a holistic framework for industrial decarbonization, developed with a major phosphate producer planning to quadruple energy consumption while transitioning to renewable sources. Our robust optimization approach combines strategic capacity expansion planning over a 25-year horizon with adaptive operational models, providing 95% reliability guarantees while balancing solar and wind integration with battery storage to meet a projected 12 TWh annual demand. In the second part, we shift our focus to developing AI systems that address the unique challenges of medical data abstraction and clinical decision support. In Chapter 4, we address the challenge of automating clinical data abstraction from electronic health records, collaborating with the Society of Thoracic Surgeons to populate their Adult Cardiac Surgery Database. Our AI pipeline combines 31 models per target variable with a two-tiered quality control system, achieving over 99% accuracy while automatically extracting 43-50% of registry variables, demonstrating how AI can dramatically reduce manual abstraction burden while maintaining clinical standards. In Chapter 5, we extend this healthcare AI focus by developing xHAIM (Explainable Holistic AI in Medicine), which addresses the limitations of current clinical AI systems in handling extensive patient records, providing interpretability, and incorporating medical knowledge. Through semantic similarity techniques and generative AI, xHAIM improves predictive performance while generating clinically grounded explanations that enhance trust and adoption by healthcare practitioners.

Multi-species genome-wide CRISPR screens identify conserved suppressors of cold-induced cell death

Mon, 01 Sep 2025 00:00:00 GMT

Multi-species genome-wide CRISPR screens identify conserved suppressors of cold-induced cell death Lam, Breanna During hibernation of Syrian hamsters, the core body temperature shows a remarkable decrease, going from 37°C to 4°C. Although this ability to survive at low temperatures could in principle be due to systemic factors that occur during hibernation, we and others have seen that cells from hibernating rodents cultured in vitro maintain this ability. Although others have studied characteristics of cells from hibernating and non-hibernating organisms, the genes and pathways that are involved in cold-induced cell death have not been systematically explored. In this thesis, we conduct two genome-wide CRISPR-Cas9 screens in both a cold-sensitive (K562) and cold-resistant (BHK-21) cell line, and uncover GPX4 and related selenocysteine incorporation genes as important for protection against cold-induced cell death. Using genetic knockdowns, along with overexpression of GPX4, we confirm our findings and demonstrate that levels of GPX4 may be limiting in K562 cells, contributing to their cold sensitivity. Additionally, pharmacological validation using inhibitors of GPX4 reveal that the catalytic activity of GPX4 is dependent on the selenocysteine in the active site. Our findings are extended across multiple cell lines and cell types across six species. Taken together, our results suggest that GPX4 may be a powerful and conserved suppressor of cold-induced cell death. Building on our initial findings, we go on to show that cold exposure leads to increases in membrane permeability. This membrane permeability is transient, as rewarming of the cells reduces permeability to baseline levels. We also test the role of lipid peroxidation in contributing to membrane permeability and find that although it contributes in some cell lines, it is not the sole contributor as ferroptosis inhibitors do not fully mitigate membrane permeability. We go on to test different membrane channels and do not see decreases in membrane permeability, potentially indicating pathway-independent effects of temperature on membrane permeability. Altogether, this work provides a foundation for understanding how cold exposure influences mammalian cells.

Why Landfills Endure: Quantifying economic barriers to material and energy recovery from municipal solid waste in the United States

Mon, 01 Sep 2025 00:00:00 GMT

Why Landfills Endure: Quantifying economic barriers to material and energy recovery from municipal solid waste in the United States Baidoo, Jacqueline E. Municipal solid waste (MSW) is a heterogeneous mixture of materials discarded by residential and nonresidential generators at end-of-life processing facilities for treatment and disposal. Conventional treatment methods reduce waste volumes through recycling via material recovery facilities, energy recovery via municipal solid waste incinerators, and biochemical conversion via composting. Even so, nearly 50% of total MSW generated in the United States was sent to landfills for final disposal in 2018 and almost half of all landfills currently in operation are expected to reach capacity by 2050. Waste planners seek to use developing resource recovery technologies like dry anaerobic digestion, gasification, and pyrolysis to narrow the gaps in end-of-life processing. Such technologies are posited to improve materials circularity and advance zero-waste landfill diversion goals by transforming residuals into electricity, fuels, and precursors to chemicals and fertilizers. However, despite demonstrated improvements to technical inefficiencies in waste valorization, numerous projects built on these technologies have failed to break through to commercial success. We investigate the contribution of regional and economic factors to the success of resource recovery projects through the lens of why landfills remain the predominant method of waste disposal. We build cost models of conventional and select developing treatment methods and use discounted cash flow analysis to estimate financial feasibility by local MSW compositions as reported in regional waste characterization studies. Findings indicate that the most critical factor to sustainable operation is consistent supply of waste materials at the quality and scale that maximize production efficiency, which is not achievable without rigorous data monitoring of MSW composition. Conversely, dependence on waste volume rather than composition makes land disposal a uniquely flexible pathway capable of subsidizing the costs of resource recovery. Progress towards landfill diversion is economically linked to the opportunity cost of avoiding landfill utilization. Unless municipalities are able to introduce subsidies elsewhere in the waste management ecosystem through gate fees and credits, projects will fail where marginal net costs of diversion exceed the revenues lost from avoided landfilling. Targeted processing of organic wastes can facilitate an average diversion of 24% for the compositions surveyed and was found to be viable for composting and dry anaerobic digestion projects at low to negligible financial losses compared to landfilling.

Enlightening Artificial Intelligence with Science

Mon, 01 Sep 2025 00:00:00 GMT

Enlightening Artificial Intelligence with Science Liu, Ziming Today’s artifciail intelligence (AI) systems, while remarkably capable, are largely black boxes. The black-box nature raises concerns for those who build AI – “How can we construct an understand AI in scientifically grounded ways?”, and those who use AI – “How can we trust systems we do not understand?”. This thesis takes a humble step towards addressing the black-box problem. Building white boxes with science (Science for AI): The prevailing paradigm in AI today – “scaling is all you need" – focuses on scaling up existing models. However, this approach often yields systems that are neither interpretable nor efficient. I argue that scientific principles offer fresh perspectives for designing more transparent and effective AI systems. This is demonstrated through Kolmogorov-Arnold Networks (KANs) inspired by mathematics, Poisson Flow Generative Models (PFGM) rooted in physical intuition, and brain-inspired modular training (BIMT) drawing insights from neuroscience, etc. Opening black boxes (Science of AI): Modern AI models exhibit a range of puzzling behaviors – such as grokking, neural scaling laws and emergent representation learning – whose underlying mechanisms remain poorly understood. I employed simplified “spherical cow” models to investigate these phenomena from the perspective of phase transitions. I will show that grokking is a special phase in the hyperparameter space, which can be controlled and eliminated. The learned algorithms after grokking also display distinct phases, called clock or pizza algorithms. AI for Science: With greater interpretability, AI systems can begin to function as “AI Scientists” capable of (re)discovering deep scientific structures from data. These include conservation laws, hidden symmetries, integrable systems, Langrangian and Hamiltonian formulations, modular structures, and high-precion solutions. I believe my research work contributes to the emerging interdiscipinary field that unites AI and Science. Building opon the foundation laid in this thesis, I envision a future in which science guides AI out of its current era of alchemy and into a true era of scientific understanding.

Quantitative modeling of 5' splice site subclass regulation and evolution

Mon, 01 Sep 2025 00:00:00 GMT

Quantitative modeling of 5' splice site subclass regulation and evolution Kenny, Connor Jens Pre-mRNA splicing is an essential molecular process required for eukaryotic gene expression. In this thesis, I present a previously unknown mechanism of splicing regulation where a family of splicing factors, the LUC7 family, compete to differentially impact 5→ splice site (5→ SS) selection in a sequence-dependent manner. I quantitatively characterize two major subclasses of 5→ SS in eukaryotes and outline distinctive features of 5→ SS in exons affected by the three human LUC7 paralogs: LUC7L2 and LUC7L enhance splicing of “right-handed” 5→ SS that exhibit stronger consensus matching on the intron side of the nearly-invariant / GU, while LUC7L3 boosts splicing of “left-handed” 5→ SS with stronger consensus matching upstream of the /GU. Using a range of experimental systems, from human cells to mutant plants, I show that LUC7 paralogs have opposing effects on these two 5→ SS subclasses and that this regulatory mechanism likely originated in the last common ancestor of animals and plants over 1.5 billion years ago. I further evaluate a competing model of 5→ SS subclass regulation involving METTL16- mediated U6 snRNA modification and reconcile both models by devising computational tools that identify sequence features predictive splicing dysregulation in transcriptome-wide datasets. Finally, I examine the evolutionary dynamics of left- and right-handed 5→ SS and propose a model of intron evolution in which codon and intron phase constraints in protein-coding genes shape both minor-to-major intron conversion and transitions between left- and right- 5→ SS subclasses.

Compiler-Hardware Co-Design for Pervasive Parallelization

Fri, 01 Sep 2023 00:00:00 GMT

Compiler-Hardware Co-Design for Pervasive Parallelization Ying, Victor A. Modern computer systems have hundreds of processor cores, so highly parallel programs are critical to achieve high performance. But parallel programming remains difficult on current systems, so many programs are still sequential. This dissertation presents new compilers and hardware architectures that can parallelize complex programs while retaining the simplicity of sequential code. Our new systems allow real-world programs to use hundreds of cores without burdening programmers with concurrency, deadlock, or data races. This dissertation follows a novel approach that eliminates the burden of explicit parallel programming to make parallel execution pervasive. This approach relies on four guiding principles. First, exploiting implicit parallelism preserves the simplicity of sequential execution. Second, dividing computation into tiny tasks, as short as tens of instructions each, unlocks plentiful fine-grained parallelism in challenging programs. Hardware-compiler co-design techniques can create many tasks in parallel and reduce per-task overheads to make tiny tasks scale to many cores. Third, new hardware and software mechanisms can compose parallelism across entire programs, removing serializing barriers to overlap executions of nested parallel subroutines. Finally, exploiting static and dynamic information for data locality reduces data movement costs while maintaining load balance on large multicore systems. This dissertation presents three systems that embody these four principles. First, T4 introduces automatic program transformations that exploit a novel hardware architecture to parallelize sequential programs. As a result, T4 scales hard-to-parallelize real-world programs to tens of cores, resulting in order-of-magnitude speedups. Second, S5 builds on T4 with novel transformations to remove needless serialization in a broad class of challenging data structures. Thus, S5 scales complex real-world programs to hundreds of cores, delivers additional order-of-magnitude speedups over T4, and outperforms manually parallelized code tuned by experts. Finally, ASH is an accelerator that demonstrates the same approach can be applied with simpler mechanisms tailored for digital circuit simulation. A small ASH implementation is 32x faster than a large multicore CPU running a state-of-the-art parallel simulator.

Optimizing Graph Neural Network Training on Large Graphs in A Distributed Setting

Thu, 01 Sep 2022 00:00:00 GMT

Optimizing Graph Neural Network Training on Large Graphs in A Distributed Setting Murzynowski, Philip Graph neural networks (GNNs) are an important class of methods for leveraging the information present in graph structures to perform various learning tasks. Distributed GNNs can improve the performance of GNN execution by dividing computation among multiple machines and scale to large graphs by partitioning graph features and the graph structure. Although distributed GNNs are able to achieve self-relative speedup, they are often slower than well-optimized code running on a single machine. For example, evaluation of the prevalent Distributed DGL system on graphs in the Open Graph Benchmark shows Distributed DGL can achieve speedup of over 2× when moving from one to four nodes, but execution of Distributed DGL on 4 nodes is 2× slower than a well-optimized GNN system, such as the SALIENT system, on a single machine. In my thesis, I argue that it is possible for a distributed GNN system to be both fast and scalable. Specifically, I show that it is possible to match the performance of well-optimized, non-distributed codes for GNN training and also achieve good scalability when running in the distributed setting. I present a system called Distributed SALIENT and motivate its design through profiling and identifying bottlenecks that arise in the distributed setting. Key components of Distributed SALIENT include the use of well-optimized code for local computations, pipelining of inter-machine communication, and a careful trade-off between data partitioning and partial replication. I evaluate Distributed SALIENT on the Open Graph Benchmark (OGB) and show that Distributed SALIENT achieves good speedup compared to SALIENT’s well-optimized single-node code while only using replication factors of roughly 5%. In fact, in experiments with training a 3-layer GraphSAGE model on the large OGB papers100M data set, Distributed SALIENT on 8 nodes is 8.6x faster than SALIENT on 1 node.

Atomistic Insights into Alloy Solidification using Machine-Learning Potentials

Mon, 01 Sep 2025 00:00:00 GMT

Atomistic Insights into Alloy Solidification using Machine-Learning Potentials Cao, Yifan Alloy solidification is a critical process in materials design and manufacturing, as it governs the formation of microstructures that determines the mechanical, thermal, and chemical properties of materials. However, direct in situ observation remains extremely challenging due the need for high spatial and temporal resolution under elevated temperatures. On the theory side, solidification is a complex phenomenon often studied using phase-field simulations, which rely on empirically fitted parameters and simplified assumptions about interfacial kinetics, limiting their predictive capability. Capturing this process at the atomistic level can yield more fundamental insights, but is hindered by the need for interatomic models that are both accurate and computationally efficient across relevant timescales and length scales. To overcome these challenges, this thesis develops and applies machine-learning interatomic potentials (MLPs) that capture the chemical complexity of metallic alloys, providing a physically accurate and computationally efficient backbone for large-scale atomistic simulations of complex alloy solidification. We first address a foundational challenge in deploying MLPs: the systematic construction of robust and transferable training datasets. Using CrCoNi as a model system, we evaluate various strategies for training MLPs to capture chemical short-range order (SRO), a critical feature in high-entropy alloys, and its effects on materials quantities of relevance for mechanical properties, such as stacking-fault energy and phase stability. It is demonstrated that energy accuracy on test sets often does not correlate with accuracy in capturing material properties, which is fundamental in enabling large-scale atomistic simulations of metallic alloys with high physical fidelity. Based on this analysis we systematically derive design principles for the rational construction of MLPs that capture SRO in the crystal and liquid phases of alloys. The resulting MLPs are validated against experimental measurements on key thermophysical properties, including melting points, heat capacities, thermal expansion coefficients, and enthalpy of SRO formation, confirming their suitability for predictive simulations. With these validated potentials, we then investigate the evolution of SRO during rapid solidification processes. Our simulations reveal that alloy processing can lead to nonequilibrium steady states of SRO that differ qualitatively from any equilibrium configuration. We attribute this behavior to an inherent ordering bias introduced by nonequilibrium dynamics during solidification. These findings suggest that conventional manufacturing processes offer new opportunities to tailor alloy properties by accessing a broader spectrum of nonequilibrium SRO states, expanding the alloy design space beyond the equilibrium spectrum. Finally, we conduct predictive solidification simulations of chemically complex alloys across experimentally relevant growth rates (0.15–2 m/s) , alloy compositions, interface orientations, and undercooling levels. These simulations capture the dynamic build up of solute partitioning at the solid-liquid interface and reveal kinetics-dependent segregation patterns that deviate markedly from equilibrium predictions. The developed framework enables direct evaluation of key kinetic properties under realistic growth conditions, including interface mobility, liquid diffusivity, and solute trapping. Altogether, this thesis develops machine-learning potentials capable of capturing the chemical complexity of metallic alloys with near DFT-level accuracy, and establishes a framework for extracting key kinetic properties through predictive simulations of alloy solidification. When combined with emerging advances in continuum-scale modeling, these results lay the groundwork for truly multiscale investigations of alloy solidification, enabling DFT-level predictive capabilities at scales directly comparable to experimental alloy design and additive manufacturing processes.

Sequential Resource Allocation and Applications in Revenue Management

Mon, 01 Sep 2025 00:00:00 GMT

Sequential Resource Allocation and Applications in Revenue Management Zhou, Zijie Sequential resource allocation is a fundamental problem in operations research, encompassing a wide range of applications where decisions must be made dynamically under uncertainty. This thesis develops new theoretical foundations, explores practical applications, and establishes evaluation methodologies for sequential resource allocation, with a focus on revenue management, robustness and fairness, and experiment design. On the theoretical side, this thesis advances the study of classical network revenue management, a long-standing challenge in dynamic resource allocation. We introduce the first LP-free algorithm, improving the regret bound from O(T ^1/2) to O(T ^3/8)—a significant step toward closing the gap between existing algorithms and the theoretical lower bound of O(1). Additionally, we enhance robustness in sequential resource allocation by developing algorithms that incorporate machine-learned advice, striking a balance between overly conservative worst-case models and overly optimistic stochastic assumptions. Furthermore, we integrate individual fairness into sequential decision-making, ensuring equitable resource allocation without compromising competitive performance. On the application side, we demonstrate the impact of sequential resource allocation in the hospitality management domain. Collaborated with Oracle Lab, we design an online upgrading mechanism that enables hotels to dynamically determine when and at what price to offer room upgrades. Additionally, we propose near-optimal, fast approximation algorithms for this mechanism, achieving a regret bound of O(logT), which is close to the natural lower bound of O(1). We also incorporate our upgrading algorithm to a hotel dataset, and improves more than 20% revenue in 2022. Finally, we introduce new methodologies for evaluating sequential decision-making policies, with a focus on online experiment design. Traditional A/B testing methods struggle with dynamically arriving data, leading to biased or inefficient experimental results. Our pigeonhole experimental design effectively reduces bias and outperforms several well-known experimental design policies, including matched pair design and completely randomized design, making it a more reliable approach for evaluating sequential decision-making strategies. By unifying theoretical insights, real-world applications, and online evaluation frameworks, this thesis contributes to the broader field of sequential resource allocation, providing fundamental advancements with practical implications across revenue management and experimental design.

Aspects of Moiré Quantum Matter

Mon, 01 Sep 2025 00:00:00 GMT

Aspects of Moiré Quantum Matter Paul, Nisarga The advent of moiré quantum matter has newly unified disparate themes in modern condensed matter physics, chief among them band theory, correlations, and topology. This thesis investigates how the interplay between these foundational elements leads to novel electronic phenomena uniquely enabled by moiré superlattices. We focus on modulated Landau levels, which is one of the simplest settings with all three of band dispersion, correlations and topology, yet is rich enough to capture much of the interesting phenomena of moiré quantum matter. We characterize emergent quantum phases that are newly unlocked by the moiré regime. Specifically, we discuss directional localization, formation of Hall crystals with tunable Chern numbers, and novel fractional Chern insulator collective mode physics in the context of modulated Landau levels. We also show that a class of models comprising itinerant electrons strongly coupled to skyrmion-like magnetic textures, closely connected with moiré transition metal dichalcogenides in which the fractional quantum anomalous Hall effect was observed, can host flat Chern bands, emergent Landau levels, and zero-field non-Abelian topological order. This thesis provides a framework for the study of the essential features of moiré quantum matter and demonstrates how moiré systems provide unprecedented opportunities to explore, design, and manipulate strongly correlated topological quantum matter.

An Optimized Bayesian Analysis Framework for the KATRIN Experiment

Mon, 01 Sep 2025 00:00:00 GMT

An Optimized Bayesian Analysis Framework for the KATRIN Experiment Xu, Weiran Neutrinos, which were originally predicted to be massless within the Standard Model of particle physics, have been confirmed to possess non-zero masses through the discovery of neutrino flavor oscillations. These oscillations precisely measure mass-squared splittings between neutrino mass eigenstates, establishing lower limits for the effective electron-neutrino mass at 0.009 eV for normal mass ordering and 0.050 eV for inverted mass ordering. However, the absolute neutrino mass scale remains a fundamental open question in both particle physics and cosmology. Precise spectroscopy of beta-decay spectrum provides a model-independent probe of the absolute neutrino mass via decay kinematics. The KArlsruhe TRItium Neutrino (KATRIN) experiment, utilizing a Magnetic Adiabatic Collimation and Electrostatic (MAC-E) filter spectrometer, sets the world's tightest upper limit of m_v < 0.45 eV (90% C.L.) based on its first five measurement campaigns. KATRIN is scheduled to complete its 1,000-day data-taking period by the end of 2025, targeting a final sensitivity of m_v < 0.3 eV}. Future improvements on neutrino mass measurements will depend on advances in differential detection techniques and the development of atomic tritium sources. This thesis presents an optimized modeling of the KATRIN beta spectrum and a comprehensive analysis of the first five measurement campaigns. An improved framework for computing the theoretical beta spectrum and the KATRIN response function is developed to address the complexities arising from the asymmetric field configurations in the main spectrometer. Benefiting from a computational speedup of four orders of magnitude and improved numerical stability, frequentist best-fit values for individual campaigns are reported, together with an upper limit on neutrino mass using the Lokhov-Tkachov confidence belt construction method. Parallel Bayesian analyses are conducted on the same dataset, yielding an independent and complementary statistical interpretation of the experimental results. Posterior distributions for the squared neutrino mass are sampled for each campaign under a flat prior on m²ᵥ using the parallel Stretch-Move algorithm, and are subsequently combined with a novel approach developed in this work to enhance computational efficiency. Convergence of each Markov chain is assessed through autocorrelation time analysis, and the robustness of the results is validated through cross-team comparison and consistency checks with profile likelihood. The Bayesian results reported here enable straightforward integration with constraints from oscillation measurements and cosmological observations, and the methodologies developed in this work are directly applicable to the final KATRIN dataset, providing a foundation for future neutrino mass analyses and searches for physics beyond the Standard Model.

Redox-Mediated Processes Toward Modular Electrochemical Systems

Mon, 01 Sep 2025 00:00:00 GMT

Redox-Mediated Processes Toward Modular Electrochemical Systems Mallia, Christopher T. Electrochemical technologies offer an attractive path toward a sustainable future where conventional methods of storing energy or producing critical materials are increasingly coupled to renewable electricity generation. To enable such a future, it is imperative that we have strong foundational understanding of electrochemical reactions that are useful to our needs. Redox flow batteries (RFBs) have emerged as a promising architecture for large scale storage of electricity to bridge the gap when renewable generation is unavailable. These devices operate by storing charge in the form of redox-active species that are dissolved into an electrolyte, and subsequently passed through an electrochemical cell to either store or release electrical energy. An extension of the concept of RFBs toward more general applications is to use the dissolved redox-active species to drive a reaction with another material, either to increase the energy storage density through an electrochemically active charge-dense material, or to drive a useful chemical reaction. This extension is termed a redox-mediated (RM) process, and inherits many of the complexities and intricacies of conventional electrochemical technologies, specifically that of RFB-type devices. The subject of this thesis is the development of knowledge and techniques for studying RM processes toward practical embodiments. While technical implementations of this concept are still nascent, many promising early results have been found in devices that use redox-mediated reactions to store electricity. Despite this, progress is frequently hindered by a lack of foundational knowledge from which to ideate better systems, and techniques to experimentally determine underlying physics. First, I establish the development of the RM concept over the past years as primarily through proof-of-concept electrochemical reactors which mimic RFBs. Second, we establish that the underlying nature of some RM reactions can be quantified and understood through corrosion principles, which guide our intuition for selecting chemistries and operating conditions. Third, I demonstrate that the behavior of many desirable RM chemistries is intrinsically coupled to passivation phenomena, and that this must be accounted for in reaction design. Fourth and finally, I provide experimental and practical guidance for researchers in this field, coupled with the design of some apparatus and techniques useful for characterizing RM reactions in specific and electrochemical processes in general. This body of work is broadly intended to advance understanding of electrochemically active interfaces and enable technology concepts which promote a sustainable future.

Predictive Modeling of Chemical Reactivity for Sustainability

Mon, 01 Sep 2025 00:00:00 GMT

Predictive Modeling of Chemical Reactivity for Sustainability Singhal, Avni Priya Predicting and controlling chemical reactivity is key to sustainable material and process design. However, modeling reactivity at scale remains challenging due to the computational demands of quantum chemical methods and the complexity of reaction mechanisms. This thesis explores how high-throughput computational approaches, rooted in quantum chemistry and enabled by automation, can be used to interrogate reactivity across large chemical spaces. We focus on two domains where reactivity governs process efficiency and sustainability: solvent-based carbon capture and polymer, specifically thermoset, manufacturing. We first investigate pi-conjugated heterocyclic nucleophiles as alternative carbon capture solvents to address the high regeneration energy and degradation rates of conventional amine-based systems. We combine synthetic template-based library enumeration, density functional theory (DFT), and machine learning models to evaluate binding energies, capture capacity, regeneration thermodynamics, and oxidative stability. Structure–property analysis reveals design strategies to enhance capture strength while balancing tradeoffs with desorption temperature and degradation resistance. We next focus on designing monomers for frontal ring-opening metathesis polymerization (FROMP), a polymerization mode that enables rapid, energy-efficient manufacturing of polymers. This self-propagating process harnesses exothermic reactions to sustain a polymerization front without continuous external heating, but it requires monomers with a finely tuned balance of thermodynamic and kinetic parameters. We develop a multi-level screening pipeline that integrates DFT-calculated properties with a reaction-diffusion model to predict front behavior directly from the atomistic structure of the monomer. We experimentally validate a preliminary pipeline, identifying a new class of FROMP-capable furan-benzyne monomers, and uncover additional candidates from unexplored chemical spaces that overcome limitations of known systems. Together, these studies demonstrate how high-throughput, mechanism-informed modeling can guide the discovery of molecules and materials that meet complex reactivity and performance criteria.

When Your Home Becomes a Panda Park: The Opportunity and Upheaval of China's Giant Panda National Park

Mon, 01 Sep 2025 00:00:00 GMT

When Your Home Becomes a Panda Park: The Opportunity and Upheaval of China's Giant Panda National Park Zhao, Celina In December 2016, China launched the Giant Panda National Park (GPNP). A massive ecological initiative aimed at safeguarding its beloved national symbol and international icon of conservation, the park marked an unequivocal win for giant pandas. But for the 100,000 people already living in and around the borders, the outcome was not as clear. The GPNP seeks to establish a harmonious balance between biodiversity protection and human development. But the vast amount of land covered by the park means not all places are equally primed to achieve that goal. A handful of communities have been designated as exclusive entrance communities, with lavish funding to become the face of the national park. In others, a persistent question simmers: Are pandas more important than people? Central to this story is how individuals are adapting to and reimagining their futures. Rather than a binary of winners and losers, the GPNP has sparked a wide range of human responses -showing that the path to a sustainable future between people and pandas is far from black and white.

The Sequence Landscape of Bacterial Genes is Shaped by Long-Range mRNA Folding

Mon, 01 Sep 2025 00:00:00 GMT

The Sequence Landscape of Bacterial Genes is Shaped by Long-Range mRNA Folding Gill, Manraj Singh An evolutionary selection for optimal expression of genes in regulatory networks has led to discernable sequence patterns in bacterial genomes observed in nature. Such patterns result from gene regulatory strategies that leverage sequence-dependent interactions with key cellular machineries and regulatory molecules. While numerous regulatory strategies that shape bacterial gene sequence have been characterized, predicting functional consequences from sequence alone remains challenging due to the sheer vastness of the possible sequence space. Moreover, the primary gene sequence encodes information on secondary and tertiary topologies that the molecules of the central dogma can fold into. Specifically, though local messenger RNA (mRNA) structures are known to regulate bacterial gene expression, the role of long-range mRNA folding remains unclear despite the predicted prevalence of such interactions across mRNAs. In bacteria, a major regulator of mRNA decay and translation rates is accessibility of the ribosome binding site (RBS) to the ribosome. Sequences in the mRNA’s 5´ untranslated region (UTR) complementary to the RBS can decrease gene expression by base pairing and occluding ribosomes from binding. To determine whether such antagonistic sequences are also the primary determinants of sequence choice along the rest of the mRNA transcript, we measured the effect of all possible 8-nucleotide substitutions (65,536 variants) on mRNA levels when placed in multiple positions along a bacterial transcript. We find that, while the vast majority of substitutions in the middle of genes negligibly affect RNA level, 8mers with complementarity to parts of the RBS exhibit the strongest effects by increasing RNA degradation rates up to 4-fold. RBS-complementary sequences also decrease translation initiation rates when placed in a coding sequence, and are able to occlude ribosome binding even when they are located hundreds of nucleotides away from the start codon. The inhibitory effect of such secondary structures on gene expression likely explains a strong selection against sequences complementary to conserved parts of RBSs throughout coding sequences of genes from diverse bacterial genomes, which we uncover through computational analysis. Together, this thesis reveals the widespread impact of RNA intramolecular interactions in vivo on both mRNA stability and translation and uncovers a key constraint on gene sequences.

Engineering Materials for Non-Compressible Torso Hemorrhage and Internal Bleeding

Sun, 01 May 2022 00:00:00 GMT

Engineering Materials for Non-Compressible Torso Hemorrhage and Internal Bleeding Hong, Celestine Jia Huey Non-compressible torso hemorrhage (NCTH) and internal bleeding results in a significant number of preventable casualties worldwide among civilians and in the field. In particular, internal bleeding can only be diagnosed through changes in vital signs and then through imaging modalities that may only be available in a hospital setting. Over the past few decades, researchers in the field have sought to address these needs by developing hemostats that can rapidly expand, bind, or seal an exposed wound, or interact with wound-specific components when delivered intravenously to enhance preexisting hemostatic processes. The first part of this thesis investigates the effect of hemostatic nanoparticle size on their interactions with platelets. Small nanoparticles were observed to result in an increased percentage of specifically-bound single platelets under flow and intermediate nanoparticles were observed to result in the greatest degree of platelet recruitment to a platelet-collagen surface. Large nanoparticles were observed to result in the most nanoparticle mass bound to a surface, the shortest circulation time and retention, and the highest pulmonary accumulation. Ultimately, intermediate nanoparticles were shown to result in the most significant increase in survival relative to the saline control in a lethal inferior vena cava (IVC) injury model (84.6% vs 26.7%), as well as the greatest accumulation at the injured IVC relative to uninjured vessel controls. Subsequently, the intermediate nanoparticles from the prior study were functionalized with bio-orthogonal click-crosslinkable azide groups to achieve targeted crosslinking behavior. Commercial multiarm PEG functionalized with the corresponding clickable moiety, dibenzylcyclooctyne (DBCO), and DBCO-PEG-b-PLGA nanoparticles were delivered as the second part of this two-component system. This system was demonstrated to increase platelet recruitment, and decrease fibrin loss during plasminolysis in vitro. When challenged in a mouse liver resection model, the two-component system resulted in significantly increased survival relative to the nanoparticle-only system and higher accumulation in the remnant liver. Finally, a charge-inverting polymer was synthesized through controlled radical polymerization. The material was demonstrated to undergo rapid charge inversion when exposed to physiological pH, resulting in the near-complete lift-off within a minute of a layer-by-layer drug film into the dermis when coated on microneedles. This versatile release platform could be coated on wound dressings to facilitate the release of therapeutics to aid in healing, or other applications involving charged films. In sum, this thesis has investigated several new materials and assays for the treatment of traumatic hemorrhage, opening potential avenues for the development of more effective hemostats.

Advances in Nonconvex and Robust Optimization

Mon, 01 Sep 2025 00:00:00 GMT

Advances in Nonconvex and Robust Optimization Koukouvinos, Theodoros Nonconvex optimization presents significant challenges, as identifying the global optimum is often difficult. This thesis introduces novel algorithms to find the exact solution of a broad class of nonconvex optimization problems. The thesis is structured into four parts. In Chapter 2, we propose a novel method for solving nonconvex optimization problems, in which the nonconvex components are sums of linear times convex (SLC) functions. We introduce a new technique, called the Reformulation-Perspectification Technique (RPT), to obtain a convex approximation of the original nonconvex optimization problem. We then incorporate RPT within branch and bound to obtain the global optimal solution of the nonconvex optimization problem. By using the RPT, we obtain a convex relaxation by forming the perspective of each convex function and linearizing all product terms with newly introduced variables. To further tighten the approximation, we pairwise multiply constraints. Therefore, in Chapter 3, we analyze all possibilities of multiplying conic constraints, a very wide class of constraints. Further, we delineate methods for deriving new, valid linear and second-order cone inequalities for pairwise constraint multiplications involving the power cone and exponential cone, thereby enhancing the strength of the approximation. In Chapter 4, we address nonconvex optimization problems that involve polynomials. We derive valid SLC decompositions for polynomials, in which the linear functions are inequalities of the feasible region and the convex functions are quadratics. We prove the existence of such SLC decompositions for arbitrary degree polynomials. Further, out of the many possible SLC decompositions, we obtain the one that results in the tightest lower bound. Finally, in the numerical experiments we show that our method often outperforms state-of-the-art approaches for polynomial optimization. In Chapter 5, we propose a robust optimization framework that immunizes some of the central linear algebra problems in the presence of data uncertainty. Namely, we formulate linear systems, matrix inversion, eigenvalues-eigenvectors and matrix factorization under uncertainty, as robust optimization problems using appropriate descriptions of uncertainty. We show that for both linear systems and matrix inversion, the robust approach leads to more accurate solutions than the nominal, in the case of nearly singular matrices.

Fabricating and Tailoring Halide Perovskites for Photovoltaic Applications

Mon, 01 Sep 2025 00:00:00 GMT

Fabricating and Tailoring Halide Perovskites for Photovoltaic Applications Kadosh Zhitomirsky, Tamar Green energy is a contemporary global concern, and research of materials for solar energy harvesting is the heart of potential solutions for the energy crisis. Halide perovskites are leading candidates to replace silicon in next generation solar cells. This thesis focuses on halide perovskite materials, aiming to understand their structure, electronic and ionic properties and photo-activity; and to re-direct their fabrication techniques to address global market needs and requirements. In this work we developed alternative, vapor-based fabrication techniques, based on manufacturing-compatible, safe, rapid and scalable processes, that have the potential to improve material stability and efficiency. Vapor Transport Deposition (VTD) is investigated as a promising fabrication method for thin film halide perovskites and beyond. We explored the deposition parameter space and elucidated relationships and trends regarding composition, structure and deposition rate. We examined the morphology, crystal phase formation, optical and electrical properties, and finally the performance of the deposited films when incorporated into solar cells. We begin by exemplifying the viability of vapor transport co-deposition in fabricating active perovskite films, utilizing methylammonium lead iodide (MAPbI3) as a simplified model system. We then design an improved version of the vapor transport deposition system and transition to the more technologically attractive perovskite composition formamidinium lead iodide (FAPbI3). Learning from previous attempts to fabricate this material, we developed a novel technique that we call Hybrid two-step vapor-solution deposition in which we use VTD to deposit the inorganic 4 precursor, not readily dissolved in industry acceptable solvents, and then react it with a solution of the organic precursors dissolved in a benign solvent. This technique allowed us to fabricate functioning FAPbI3 based solar cell devices, in a safe, fast-paced, scalable and manufacturing compatible fashion. The deposition rate is significantly influenced by chamber pressure and source temperature, and by controlling all deposition parameters, we systematically reached rates of up to 1200 nm/min, that is orders of magnitude faster than current comparable techniques. We found the technique to be reproducible, yielding 13% efficient devices, with champion efficiencies of up to 15.3%. Based on the proposed novel fabrication process, we believe it offers an avenue for further improvement in solar cell stability and efficiency. CsPbBr3, a fully inorganic halide perovskite, also shows great promise as a photo and gamma ray detector and like the other halide perovskites is known to support halide ion conductivity that contributes to device instability and reduced sensitivity to irradiation. We choose this as a model system to apply concepts from defect chemistry and demonstrate the ability to measure and manipulate the ionic conductivity in the material by stoichiometry control and doping.

Membrane protein conformational dynamics and ligand-binding interactions in bacterial glycoconjugate biosynthesis

Mon, 01 Sep 2025 00:00:00 GMT

Membrane protein conformational dynamics and ligand-binding interactions in bacterial glycoconjugate biosynthesis Higinbotham, Hugh Membrane associated proteins are an essential component of the complex biochemistry that is carried out at the membrane interface and perform essential functions for cellular life. Biophysical characterization of protein structure-function relationships faces a unique set of challenges due to the constraints of phospholipid bilayer chemistry and geometry. Advances in x-ray crystallography and cryo electron microscopy have made progress in this regard, but dynamic structural features remain difficult to study. Small membrane proteins, such as those responsible for bacterial glycosylation, remain challenging to structurally characterize at all. Bacterial glycan synthesis pathways are essential for cell function yet highly variable between strains, making them promising systems for targeted antibiotic development. Many pathways have initiating SmPGTs that show incredible specificity for minute changes in glycan chemistry despite being small enough to streamline many computational methods, which makes them ideal model systems for developing multidisciplinary strategies to study membrane protein dynamics. This thesis presents a strategy that employs structural bioinformatics in Chapter 2, molecular dynamics simulation (MD) in Chapter 3, and single-molecule FRET microscopy (smFRET) in Chapter 4 to observe the ligand-dependent conformational dynamics of integral membrane proteins in situ. It focuses on representative members of the small monotopic phosphoglycosyl transferase (SmPGT) superfamily, which catalyze transfer of a phosphosugar from a soluble nucleotide-sugar donor to a membrane-embedded polyprenol phosphate acceptor in the initiating step of glycoconjugate biosynthesis in prokaryotes. The pipeline is employed to confirm the role of SmPGT conformational dynamics in substrate binding and informs the design of non-hydrolyzable substratemimetic inhibitors. Chapter 5 further sets the stage for the use of structural bioinformatics and molecular simulation to characterize subsequent glycosyl transferase (GT) enzymes down pathway and presents initial results characterizing inter-protein cooperative interactions. The integrated approach to incorporate computational and experimental characterization methods has significantly contributed to the understanding of SmPGT structure-function relationships and opened up new directions of inquiry into specific PGTligand interactions, the development of new inhibitory compounds, and the role of interprotein interactions in bacterial glycan synthesis.

Functional Genomic and Image-Based Screening Approaches for Probing Host-Pathogen Interactions

Thu, 01 Jun 2023 00:00:00 GMT

Functional Genomic and Image-Based Screening Approaches for Probing Host-Pathogen Interactions Carlson, Rebecca J. Host-pathogen interactions represent a complex interplay between hosts and pathogens that can evolve over millions of years. Interactions between bacteria or viruses and human cells, and the resulting evolved antipathogenic signaling pathways, are processes responsible for pathologies ranging from infectious diseases to autoimmune conditions and cancer. In addition, engineered designs inspired by pathogen interactions with hosts are increasingly being used to both treat and diagnose many pathologies that need not originate from infection with a pathogen. Therefore, it is critical to build and deploy scalable tools to better understand host-pathogen dynamics in order to both better treat conditions where pathogens or antipathogenic signaling contribute directly to disease pathology as well as to engineer new treatments to address a broader range of disease states. In this thesis, I describe approaches to leverage functional genomics and image-based screening to perturb and profile host-pathogen interactions, including responses to two RNA viruses, Sendai virus and Ebola virus. These provide case studies highlighting the utility of high-content image-based screening for revealing new genes regulating predefined phenotypes of interest as well as for generating single-cell imaging profiles that can be used to infer new genetic functions and phenotypic states directly from screening data without a priori specification. I also highlight an example of a genetic screen that revealed a robust negative result, leading to hypothesis and validation of a novel function of the STING protein as a proton channel.

A study of the field control operation of railway motors : a thesis

Thu, 01 Jan 1925 00:00:00 GMT

A study of the field control operation of railway motors : a thesis Davis, Stanley W. (Stanley Whitcomb) Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1925; Includes bibliographical references (leaf 91).

Reviewing I.S. : how to handle legacy systems?

Fri, 01 Jan 1999 00:00:00 GMT

Reviewing I.S. : how to handle legacy systems? Orlando, Ricardo, 1966- Thesis: S.M.M.O.T., Massachusetts Institute of Technology, Sloan School of Management, Management of Technology Program, 1999; Includes bibliographical references (leaves 100-106).

The effects of changing economic conditions on energy costs in stainless-steel clad pressurized water reactors

Mon, 01 Jan 1962 00:00:00 GMT

The effects of changing economic conditions on energy costs in stainless-steel clad pressurized water reactors Trapp, Donald L. Thesis: M.S., Massachusetts Institute of Technology, Department of Nuclear Engineering, 1962; Appendix contains numerous pamphlets.; Includes bibliographical references (leaves 135-136).

The politics of metropolitan transportation.

Wed, 01 Jan 1964 00:00:00 GMT

The politics of metropolitan transportation. Colcord, Frank Carlton. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Economics and Social Science, 1964

The design of a control system for the terminal phase of a satellite rendezvous

Thu, 01 Jan 1959 00:00:00 GMT

The design of a control system for the terminal phase of a satellite rendezvous Hollister, Walter M., 1930- Thesis: M.S., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 1959; Includes bibliographical references (leaf 47).

Compliance in a gyroscope gimbal

Wed, 01 Jan 1958 00:00:00 GMT

Compliance in a gyroscope gimbal Graham, James William. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1958

On hypergraphs and hypergeometries.

Fri, 01 Jan 1971 00:00:00 GMT

On hypergraphs and hypergeometries. Helgason, Thorkell. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Mathematics, 1971; Vita.; Bibliography: leaves 158-159.

Noise analysis of circuit models representing maser operation.

Fri, 01 Jan 1965 00:00:00 GMT

Noise analysis of circuit models representing maser operation. Hempstead, Robert Douglas. Thesis: M.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1965; Bibliography: leaves 106-108.

Cocoa in the Ghanaian economy.

Fri, 01 Jan 1965 00:00:00 GMT

Cocoa in the Ghanaian economy. Bateman, Merril Joseph. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Economics, 1965

Even denominator quantum numbers and termination of the fractional series in the fractional quantum hall effect

Sun, 01 Jan 1989 00:00:00 GMT

Even denominator quantum numbers and termination of the fractional series in the fractional quantum hall effect Willett, Robert L. (Robert Lee) Thesis: Ph. D., Massachusetts Institute of Technology, Department of Physics, 1989; Includes bibliographical references (leaves 6-7).

Improving railroad terminal control systems : a case study of Southern Railway's Brosnan yard

Mon, 01 Jan 1979 00:00:00 GMT

Improving railroad terminal control systems : a case study of Southern Railway's Brosnan yard Ferguson, William Lloyd. Thesis: M.S., Massachusetts Institute of Technology, Department of Civil Engineering, 1979; Bibliography: leaves 194-195.

Investigations in the theory of quantum corrections to classical solutions of the Yang-Mills equations

Mon, 01 Jan 1979 00:00:00 GMT

Investigations in the theory of quantum corrections to classical solutions of the Yang-Mills equations Callias, Constantine John. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Physics, 1979; Includes bibliographical references.

Wave equations, particles and chronometric geometry.

Thu, 01 Jan 1976 00:00:00 GMT

Wave equations, particles and chronometric geometry. Orsted, Bent. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Mathematics, 1976; Includes bibliographical references.

Planteez(tm) : business plan and preliminary research

Mon, 01 Jan 2001 00:00:00 GMT

Planteez(tm) : business plan and preliminary research Sanchez, Manuel A. (Manuel Andres), 1979- Thesis: S.B., Massachusetts Institute of Technology, Department of Mechanical Engineering, 2001; Includes bibliographical references (p. 15).

Systematic engineering of controlled, localized oligonucleotide delivery systems for wound angiogenesis

Wed, 01 May 2024 00:00:00 GMT

Systematic engineering of controlled, localized oligonucleotide delivery systems for wound angiogenesis Berger, Adam G. The standard of care for diabetic wounds has remained relatively unchanged for decades, resulting in patients with wounds that do not heal on meaningful time scales, referred to as ulcers, and high rates of recurrence for patients whose wounds do heal. This common complication of diabetes decreases quality of life, increases mortality, and raises health care costs. New paradigms to treat these wounds remains a formidable but critical challenge. Addressing diabetic ulcers at the molecular level may decrease healing time and prevent recurrence. Impaired blood vessel formation, or angiogenesis, in diabetic ulcers is an important target pathway. Angiogenesis is needed to bring oxygen, nutrients, signaling cues, and cells to newly formed tissue while removing waste. Nucleic acid oligonucleotide therapies, such as small interfering RNAs (siRNAs) or microRNA inhibitors (anti-miRs), that regulate gene expression at the post-transcriptional level, hold particular promise for promoting angiogenesis and wound healing; however, the large size and negative charge of these therapies require drug carriers to mediate their biological effect. In this thesis, we leverage sequential electrostatic adsorption of oligonucleotide therapy and polyelectrolytes into thin film coatings on commercial wound dressings through the layer-by-layer (LbL) process. These dressings package oligonucleotide, enhance its transfection efficacy, and control its temporal release locally to the wound bed. After initial validation experiments, we sought to systematically understand our drug carrier system and use this insight to engineer better wound dressings. First, we developed a proof-of-concept anti-miR-coated dressing and showed its efficacy in promoting both wound closure and sex-dependent angiogenesis. We found that therapy released from coated dressings had a preferential association with different wound cell types, particularly endothelial cells. We then sought to uncover how changes in the oligonucleotide structure itself may alter interactions with transfection polymers in thin film coatings. We found that binding with certain polyelectrolytes differed based on whether the therapy was a flexible single stranded anti-miR or a more rigid double stranded helix siRNA. We also showed how chemically modified nucleotides, such as locked nucleic acid and 2’-O-methyl RNA, can modulate affinity to polyelectrolytes and ultimately impact transfection efficacy. We also elucidated how physicochemical properties of the hydrolysable transfection-enhancing poly(β-aminoester) polymer mediate its efficiency in transfecting oligonucleotide therapy. We demonstrated that a more hydrophobic polymer enhanced transfection efficacy through its ability to facilitate permeation of biological barriers. Finally, we identified how modulation of the anionic excipients contained in these thin film coatings can be leveraged to vary the release kinetics from coated wound dressings. We engineered formulations that released on a fast or slow time scale. We observed that while both release time scales promoted efficacy in wound closure, they did so through potentially different mechanisms despite the same putative pro-angiogenic anti-miR therapy. In sum, this thesis elucidates how physicochemical properties and formulation of coated wound dressings alter their interfacial effects with biological systems. We use this knowledge to rationally design better drug carriers that can deliver pro-angiogenic oligonucleotide therapeutics to the wound bed. The findings have broad applications in the delivery of nucleic acid therapies for a wide host of diseases where local delivery to the injured tissue could prove beneficial. Ultimately, we also advance our pro-angiogenic coated wound dressing strategy towards clinical translation. Our strategy has the potential to provide a new, targeted therapeutic paradigm to help those suffering from diabetic ulcers.

Atmospheric Impacts of Hydrogen as an Aviation Fuel

Wed, 01 May 2024 00:00:00 GMT

Atmospheric Impacts of Hydrogen as an Aviation Fuel Gibney, Evan M. Hydrogen is being investigated as a promising zero-carbon aviation fuel, offering the potential to eliminate direct CO₂ emissions while being produced with low lifecycle greenhouse gas emissions. Despite these benefits, there are additional indirect climate and air quality costs associated with direct hydrogen emissions which are often overlooked. We quantify the perturbation in the atmospheric composition associated with the introduction of hydrogen-fueled aircraft, broadening the current understanding of the non-CO₂ effects of these fleets. We use the GEOS-Chem High Performance (GCHP) global chemistry-transport model to conduct a spatially discretized, multi-year impact assessment of the atmospheric impacts of hydrogen-fueled aviation. We implement a flux surface boundary condition for hydrogen to provide an improved representation of the soil sink, relative to the default fixed boundary condition. This results in a net surface exchange of-16.7 Tg H₂ per year. Two hydrogen scenarios are evaluated using the updated GCHP implementation, which are representative of a high and low mitigation scenario for direct hydrogen emission rates. For the two scenarios, respectively, we observe increases in the mean atmospheric methane mixing ratio of 3.34 ppbv and 10.7 ppbv, corresponding to an increased methane lifetime of between 0.24% and 0.77%, respectively. The increased methane lifetime as well as in-situ oxidation of stratospheric hydrogen results in an increased stratospheric water vapor burden of 0.42 Tg and 2.3 Tg (or 0.052% and 0.28%) for the high and low mitigation scenarios, respectively. Additionally, we show the perturbation to tropospheric ozone levels to be between-0.047% and +0.30%, where the decreased ozone results from the removal of NOₓ emissions associated with fuel cells and low hydrogen emission rates. This analysis provides the foundation for understanding the implications of potential future hydrogen-based aviation fleets on climate and air quality.

A Changing Climate Beneath Our Feet: How plant and microbial life in tropical soils are shifting and what that could mean for the future of our warming planet

Mon, 01 Sep 2025 00:00:00 GMT

A Changing Climate Beneath Our Feet: How plant and microbial life in tropical soils are shifting and what that could mean for the future of our warming planet Ocharoenchai, Nanticha Discussions about climate change and carbon sequestration have largely revolved around plant structures we can easily see, like leaves that absorb CO₂ for photosynthesis and woody trunks that store carbon as biomass. Carbon credits that companies and consumers buy to compensate for emissions they’ve produced are primarily calculated based on these parts, as are models that predict climate change impacts. But researchers are now beginning to understand that what we see aboveground is only part of the equation. The other part lies beneath our feet in an intricate, expansive, covert realm where plant roots, microbial communities and soil dynamics interact. These belowground systems are crucial for cycling carbon through the Earth and regulating the climate, but relatively little is known about them compared to aboveground systems. This is especially true in tropical regions, where one-third of the world’s terrestrial carbon storage lies. However, these systems are evolving quickly with climate change, contradicting what models have previously projected. With so many global decisions based on such models, these uncertainties hold planetary significance for our future. A group of scientists is climbing an uphill battle, racing against time to understand this understudied field.

Engineering Winter

Mon, 01 Sep 2025 00:00:00 GMT

Engineering Winter White, Mackenzie As winters warm and snowfall becomes less reliable, ski resorts worldwide increasingly depend on artificial snow to stay open. Snowmaking, once a stopgap, has become the backbone of entire seasons in a sprawling choreography of pumps and pressurized mist designed to hold trails together. At resorts like Vermont’s Bromley Mountain, snowmakers work through the night, drawing millions of gallons from limited reservoirs and operating within narrowing windows of cold air. What emerges is a portrait of winter in transition: less predictable, more expensive, increasingly manufactured. The efforts to preserve winter recreation carry growing costs in energy, water, and equitable access. Many smaller, independent ski areas struggle to meet the demands of climate adaptation, while larger resorts expand their operations, widening the divide in who can afford to sustain operations. In the American West, where rivers depend heavily on snowpack melt, the spread of snowmaking ties winter recreation to a water system already under immense strain. As artificial snow becomes the norm, winter is increasingly a season bought, built, and rationed, raising the question of whether attempts to keep the season alive are accelerating the changes that threaten to erase it.

Influence of Electronic Structure and Lattice Dynamics on Oxygen Ion Transport in Solid-State Ionic Conductors

Mon, 01 Sep 2025 00:00:00 GMT

Influence of Electronic Structure and Lattice Dynamics on Oxygen Ion Transport in Solid-State Ionic Conductors Vivona, Daniele Solid-state oxygen ion conductors are crucial for electrochemical devices such as separation membranes, solid-oxide electrolyzers, fuel cells, and sensors, serving as a technological link between renewable energy generation and consumption. Currently, these conductors are limited by slow transport rates and high operational temperatures, which pose challenges and increase costs. Developing faster conductors that operate at lower temperatures requires reducing activation energy and enhancing the pre-exponential factor in the Arrhenius equation of conductivity. However, our understanding of the fundamental processes in oxygen ion transport and methods to improve oxygen ion conductivity remain limited. This thesis focuses on understanding the fundamental mechanisms that regulate oxygen ion transport. First, the migration energy barrier in perovskite oxides is linked to an electronic energy penalty from local charge screening near the hopping ion. The energy of local electronic states is identified as a fundamental descriptor of the migration barrier. Next, migration entropy and phonon density of states (DOS) are highlighted as the main factors regulating the pre-exponential factor of oxygen ion conductivity across different materials. The phonons of oxygen ions near the hopping ion significantly contribute to migration entropy, suggesting that migration entropy can be tuned by designing the phonon dynamics of these atoms. These results imply that a widely observed correlation between increasing pre-exponential factors and activation energy arises from coupling local electronic energy states and phonons. The results are extended to the formation of oxygen vacancies and interstitials in perovskite and RuddlesdenPopper oxides. We find that defect formation energy rises with defect formation entropy, which is linked to electronic energy states interacting with phonons. In perovskite oxides, lower vacancy formation entropy is correlated with increasing oxygen phonon band center and shortening bond lengths with oxygen vacancy formation. In Ruddlesden-Popper oxides, lower interstitial formation entropy is associated with reduced octahedral tilting and local phonon changes. This thesis establishes a theoretical foundation for treating migration entropy and defect formation entropy as design variables in next-generation ionic conductors. By highlighting the impact of electronic structure and lattice dynamics on energy barriers and entropic drivers, the findings suggest new pathways for material design through the strategic separation of these factors and the intelligent design of lattice moieties in oxygen ion transport environments.

IP Networks Over Heterogeneous Embedded Serial Links

Thu, 01 May 2025 00:00:00 GMT

IP Networks Over Heterogeneous Embedded Serial Links Perry, Nathan The Internet Protocol (IP) provides a number of key benefits to networked devices: it serves as a "narrow waist" enabling functional modularity by decoupling lower-layer devices from application behavior, it provides a notion of transitive connectivity and a number of standardized methods to achieve it, and most importantly, it is ubiquitous, enabling almost all networked applications to mutually communicate. Many embedded microcontrollers cannot take advantage of the benefits of IP because they lack the dedicated networking hardware which is as a practical matter required to interact with nontrivial networks. I observe that multihop point-to-point IP networks can in principle be constructed over the communication media that microcontrollers commonly do have, such as UARTs, I2C, SPI, and CAN bus, but software support is lacking to make this networking approach accessible. Therefore, this thesis develops and evaluates interstice, a platform-independent, open-source software library designed to enable the flexible implementation of modular packet forwarders in userspace. It can be used to interconnect devices and their IP stacks across a variety of conventional and unconventional links. Interstice exposes a reprogrammable, dynamically-updatable packet-forwarding strategy, enabling forwarder nodes in principle to act as hubs, bridges, full routers, or implement firewalls or NAT, as application requirements and platform constraints permit. This approach enables benefits for modular, networked systems of microcontrollers which need to talk to the outside world: using IP enables internal microcontrollers to communicate with external devices such as PCs and smartphones without the need for application gateways. Further, to the extent that such networks are runtime-reconfigurable, features of IP such as address assignment, dynamic routing, and link-agnosticity can be incredibly beneficial. Interstice is evaluated here primarily against networks of various types of serial links (UART, I2c, CAN) speaking PPP, selected to demonstrate utility of the approach to connect embedded devices lacking dedicated networking peripherals, and further that link drivers can be specialized to take advantage of the specific characteristics of each link. The approach is showcased in application scenarios including a networked milling machine, and is analyzed for a number of performance metrics.

BioLIG: Designing Biologically Derived Electronics and Their Speculative Lives

Thu, 01 May 2025 00:00:00 GMT

BioLIG: Designing Biologically Derived Electronics and Their Speculative Lives Li, Yuqing Lucy Imagination is the origin of reality. Cultivating new infrastructural and ecological imaginaries is crucial to addressing the climate crisis. Where is the space to prototype new social and technological relations? Transient electronics is an emerging field in advanced materials focused on making electronics that don’t last. Devices are designed to be transient for biomedical, environmental monitoring, or energy storage applications. It is a fascinating and unconventional direction that advances the area of biocompatibility, redefining waste and time-programmable decay {Making electronics that, 2022}. However, in a manufacturing system that fundamentally favors the inert and invariant, transient properties can be precisely the qualities that make adaptation most challenging, often failing at the very stage of imagination. Taking inspiration from transient electronics, this thesis consists of a set of novel biomaterials, a workflow, and three fictional stories to enrich our imagination and instill agency amidst entangled humanitarian, ecological, and technological crises. BioLIG is a material for prototyping accessible and compostable electronics. It uses laser-induced graphene as an organic, bio-derived conductor and affordable biomaterials as the substrate. Three sheets and two inks make up a toolkit to create biocomposites with different properties, colors, and textures specifically designed for prototyping sensors and circuits with transient behaviours. Through a series of characterisations, BioLIG is evaluated and demonstrates that with one material, its electrical performance is on par with synthetic substrates. However, the goal is not to create a replacement material but to prototype new social and technological relations to transient materials. Through a questionnaire, I collected stories, ideas, and questions from makers, designers, and artists for BioLIG and used those as the basis for imagination. In a speculative house, on three floors, three stories unfold of a hoarder, a city forester, and a family living in a time with a leap in our relationship to fabrication, to electronics, and to decay.

Being. Creative. Together. Designing Technologies That Center Human Connection, Co-creativity, and Calm in the Era of AI

Thu, 01 May 2025 00:00:00 GMT

Being. Creative. Together. Designing Technologies That Center Human Connection, Co-creativity, and Calm in the Era of AI Dhariwal, Manuj As Artificial Intelligence (AI) becomes increasingly interwoven into our creative, social, and learning experiences, we must ask: Will these technologies deepen our connection to the timeless human experiences of Being, Being Together, and Being Creative Together—or will they pull us apart, leaving us more anxious and isolated? In an era where AI systems are increasingly framed as our “co-creators” and “companions,” enabling hyper-personalized yet hyper-isolated interactions, this dissertation reclaims the prefix ‘co-’ as fundamentally interhuman—introducing a set of new paradigms that center human connection, co-creativity, and calm in the design of technologies. Central to this work, we’ve developed CoCo (coco.build), a general-purpose, real-time co-creative learning platform that empowers young people to engage in a wide variety of safe, shared creative experiences with their peers—spanning creative computing, AI education, digital art, writing, and more. Through the platform, we showcase how digital environments can move beyond isolated modes of learning and creating to support multiple ways of being creative together with others—introducing a new paradigm for real-time digital collaboration. We further illuminate how CoCo has been envisioned as a “self-less” social platform that de-emphasizes comparison-based, self-centric metrics (profiles, likes, followers) prevalent in most online systems for young people. We weave these interconnected ideas into the unifying theme of “Being. Creative. Together.”— values we believe are both timeless and especially timely in the AI era. We supplement the broader design, technical, practical, and pedagogical contributions of this work by sharing insights and feedback from pilots with over 2,000 young people and educators across diverse settings. Ultimately, we see this dissertation as both a contribution and a call—to preserve the human essence of co-, to distinguish it from the useful, powerful, but instrumental AI interactions, and to shape digital environments that nurture our capacity to co-imagine, co-create, co-learn, co-exist, and co-evolve—with and through one another. Note: This work has been co-developed with Shruti Dhariwal. See https://coco.build/thesis for suggested citation and updates on this work.

Toward the computational transformation of legal theory and practice

Thu, 01 May 2025 00:00:00 GMT

Toward the computational transformation of legal theory and practice Mahari, Robert This doctoral thesis seeks to advance the formalization of computational law as a distinct research discipline. It explores three interwoven key themes: the empirical understanding of legal systems through advanced computational methods; the development of computational tools to augment the capabilities of legal practitioners, thereby expanding access to justice; and the identification of novel, computationally-enabled regulatory interventions. This research directly confronts the global access to justice crisis and the shortcomings of conventional legal services that frequently leave businesses and individuals without adequate support. Furthermore, the thesis investigates innovative regulatory strategies for emerging technologies, aiming to synchronize legal frameworks with contemporary technological progress by exploring adaptive and forward-looking governance approaches.

Modular Development Platforms and Creative Ecosystems: Design & Deployment for Wide Impact Across Fields

Thu, 01 May 2025 00:00:00 GMT

Modular Development Platforms and Creative Ecosystems: Design & Deployment for Wide Impact Across Fields Shtarbanov, Ali Physical, digital, and conceptual tools and building blocks are fundamental enablers and accelerators of humanity’s progress in technology, science, medicine, art, and even in abstract fields like mathematics, philosophy, and social sciences. Hardware development platforms present a special class of tools and building blocks, facilitating and accelerating innovation, prototyping, and research. They drastically reduce prototyping time and complexity, improve efficiency for experts, democratize access to innovation, and even inspire entirely new ideas. This research investigates how to design, develop, and deploy development platforms in ways that maximize their real-world impact potential. It focuses not only on the technical and engineering aspects, but also on the complete ecosystem a platform needs in order to have impact, including community building, engagement with users and volunteers, content strategy, online presence, publicity, deployment, feedback loops modularity, financial viability, and symbiotic relationships. A comprehensive Design & Deployment Framework is introduced as a conceptual tool for creating high-impact platforms and creative ecosystems, recognizing and fostering the positive feedback loops that sustain them and that shape their evolution and growth. This framework is applied in the development and deployment of multiple novel platform and ecosystem projects, including FlowIO, SleeveIO, and ModiStrap, as well as the ecosystem SoftRobotics.IO. Those works have benefited thousands of people around the world, providing researchers, designers, and engineers with powerful, reconfigurable, modular enabling artifacts that streamline prototyping, accelerate research, and lower barriers in fields like soft robotics, haptics, assistive technology, shape-changing interfaces, interactive arts, and more. A multitude of research, art, and engineering projects made possible by FlowIO and SoftRobotics.IO are presented, as well as over a dozen case studies showcasing how other users across disciplines have adopted, utilized, and extended these systems to advance their own creative, educational, and technical endeavors. Additionally, this thesis also investigates various deployment models for hardware and introduces a new hardware deployment model for equitable access to expensive hardware that may otherwise be financially out of reach for many users, as well as an “earned open-source” model, which preserves the essence of the traditional open-source model, while eliminating many of its pitfalls.

Advancing Biosecurity in the Age of AI: Integrating Novel Detection, Suppression, and Evaluation Approaches

Thu, 01 May 2025 00:00:00 GMT

Advancing Biosecurity in the Age of AI: Integrating Novel Detection, Suppression, and Evaluation Approaches Justen, Lennart J. Civilization confronts a growing challenge: advancing transformative biological science while safeguarding against catastrophic misuse, a tension amplified by the rapid convergence between biology and artificial intelligence. The COVID-19 pandemic starkly revealed our vulnerabilities to self-replicating, exponential biological phenomena, yet current defenses remain dangerously inadequate—often blind to novel pathogens until too late and lacking barriers against rapid airborne transmission. This thesis argues that robust biosecurity enables, rather than hinders, progress, and advances three key defensive capabilities. First, it evaluates blood metagenomics for pathogen-agnostic surveillance, reanalyzing public datasets to quantify viral signatures and guide the implementation of much-needed early-warning systems sensitive to novel pathogens. Second, it advances far-UVC, a type of ultraviolet between 200-235 nm, for continuous indoor air disinfection, critically assessing its safety profile through an international expert review and establishing research priorities essential for deploying this vital physical defense against airborne threats. Third, it develops rigorous methodologies for evaluating AI's rapidly evolving biological capabilities, benchmarking frontier models across diverse tasks to track progress, reveal limitations in current assessments, and guide responsible innovation in this powerful dual-use technology. Collectively, these contributions help accelerate technologies to mitigate biological risks, thereby helping secure the conditions for continued, beneficial advancement of biology in the age of AI.

From Dialogue to Decision: An LLM-Powered Framework for Analyzing Collective Idea Evolution and Voting Dynamics in Deliberative Assemblies

Thu, 01 May 2025 00:00:00 GMT

From Dialogue to Decision: An LLM-Powered Framework for Analyzing Collective Idea Evolution and Voting Dynamics in Deliberative Assemblies Poole-Dayan, Elinor Deliberative assemblies—representative samples of citizens engaged in collective decision-making through facilitated learning and deliberation—are increasingly recognized as powerful tools for revitalizing democratic governance. Yet, core aspects of how deliberation shapes which ideas advance, how perspectives evolve, and why certain recommendations succeed remain opaque and underexamined. This thesis addresses these gaps by investigating: (1) How might we trace the evolution and distillation of ideas into concrete recommendations within deliberative assemblies? and (2) How does the deliberative process shape delegate perspectives and influence voting dynamics over the course of the assembly? To answer these questions, I develop LLM-based methodologies for empirically analyzing transcripts from a tech-enhanced student deliberative assembly. The first framework identifies and visualizes the space of expressed suggestions, revealing that seemingly large gaps between ideas and final recommendations often reflect productive deliberative filtering—while also surfacing overlooked viable ideas. A second analysis integrates post-assembly survey data with transcript-grounded voting patterns to uncover the primary drivers of vote change: edits to recommendations, evolving opinions, and strategic shifts in response to updated priorities. Building on this, I introduce a framework for reconstructing each delegate’s evolving stance across the assembly, linking shifts in perspective to specific deliberative moments and justifications. Together, these methods contribute novel empirical insight into deliberative processes and demonstrate how LLMs can surface high-resolution dynamics otherwise invisible in traditional assembly outputs. The findings lay groundwork for new tools that support facilitators and delegates during live assemblies, improve transparency for decision-makers, and elevate ideas that may otherwise be missed. Looking ahead, this work opens pathways for comparative research across assemblies and highlights the potential for human-centered AI to meaningfully enhance deliberative democratic practice. As societies seek new modes of participatory governance amid growing polarization and institutional mistrust, tools that strengthen deliberation without compromising its core human character are urgently needed.

Private, Verifiable, and Auditable AI Systems

Thu, 01 May 2025 00:00:00 GMT

Private, Verifiable, and Auditable AI Systems South, Tobin The growing societal reliance on artificial intelligence necessitates robust frameworks for ensuring its security, accountability, and trustworthiness. This thesis addresses the complex interplay between privacy, verifiability, and auditability in modern AI, particularly in foundation models. It argues that technical solutions that integrate these elements are critical for responsible AI innovation. Drawing from international policy contributions and technical research to identify key risks in the AI pipeline, this work introduces novel technical solutions for critical privacy and verifiability challenges. Specifically, the research introduces techniques for enabling verifiable and auditable claims about AI systems using zero-knowledge cryptography; utilizing secure multi-party computation and trusted execution environments for auditable, confidential deployment of large language models and information retrieval; and implementing enhanced delegation mechanisms, credentialing systems, and access controls to secure interactions with autonomous and multi-agent AI systems. Synthesizing these technical advancements, this dissertation presents a cohesive perspective on balancing privacy, verifiability, and auditability in foundation model-based AI systems, offering practical blueprints for system designers and informing policy discussions on AI safety and governance.

Language Models as Mirrors and Bridges for Intergroup Communication

Thu, 01 May 2025 00:00:00 GMT

Language Models as Mirrors and Bridges for Intergroup Communication Jiang, Hang This dissertation explores how large language models (LLMs) can serve dual roles in intergroup communication: as mirrors that reflect intergroup differences, and as bridges that facilitate communication across group boundaries. Intergroup communication refers to interactions between individuals from different social groups, such as political, cultural, or professional communities, where divergent perspectives often lead to misunderstandings, unequal access to information, and social fragmentation. The first part of the dissertation presents LLMs as mirrors that reveal intergroup differences. We first introduce CommunityLM, a novel framework for probing public opinion by fine-tuning LLMs on social media posts from specific communities. Our case study comparing Republican and Democratic groups reveals that model predictions align well with human survey responses, substantially outperforming established baselines. Building on this foundation, we develop PersonaLLM to investigate whether prompt-based LLM agents can generate content aligned with assigned personas, which has emerged as a popular approach for modeling the behaviors of social groups. Through automated and human evaluations, we demonstrate that these agents can complete personality tests and write stories that reflect the distinctive behavioral patterns of specific personality profiles. Together, these complementary projects illustrate how LLMs can effectively capture and simulate the unique perspectives and behaviors that characterize diverse social groups. The second part of the dissertation presents LLMs as bridges that facilitate communication across group boundaries. First, we introduce Bridging Dictionary, an interactive tool that uses retrieval-augmented generation (RAG) techniques with LLMs to identify polarized language and suggest more inclusive alternatives. In collaboration with PBS Frontline, we demonstrate the potential of LLMs to reduce misunderstanding in journalism and political communication. Second, we present Legal Storytelling, a human-LLM collaboration framework that generates accessible narratives to explain complex legal concepts to non-experts. Through randomized controlled trials (RCTs), we find that LLM-generated narratives can improve legal literacy and help bridge communication gaps between experts and laypeople, particularly among non-native English speakers. Third, we develop FaciliTrain, a voice-based, LLM-powered system that enables facilitators to learn and practice intergroup dialogue skills with multiple LLM agents representing diverse social backgrounds and personas in a small-group setting. User studies with campus participants show encouraging early results, suggesting that LLMs can effectively support the development of communication skills essential for constructive intergroup dialogue. Together, these projects illustrate how LLMs can actively foster mutual understanding across social divides by promoting more inclusive, accessible, and constructive communication.

Delibrary: From Discussion to Outcomes and Back(casted) Again, a Visualization Tool for Deliberative Assemblies

Thu, 01 May 2025 00:00:00 GMT

Delibrary: From Discussion to Outcomes and Back(casted) Again, a Visualization Tool for Deliberative Assemblies Wong, Wing Cheung Michael With trust in traditional democratic institutions waning, it is increasingly important to examine how potential new institutions could be created and bolstered, with particular emphasis on restoring trust and empowering the public. One potential solution, the citizen's or deliberative assembly, can serve to bridge the governance and legitimacy gap between real-world policy decision-making processes and citizen-driven impact by leveraging random sortition and a well-designed deliberation process. In this thesis, I explore how AI-driven sensemaking via GPT4o-mini--a Large-Language Model (LLM)--synthesized with custom-built visualization tools, can potentially reveal the dynamics within citizen deliberative assemblies where representative, randomly selected citizens navigate public interest issues through facilitated deliberation--and how such tools can serve to amplify transparency within both the assembly process itself and the issues they explore. Through building three different prototype visualization frameworks and the development of an AI-powered topic identification process called backcasting, I analyze novel datasets from two tech-enhanced assemblies; fully recorded discussions from both an on-the-ground citizens' assembly in Deschutes County, Oregon, as well as an MIT student assembly on sustainability. In backcasting, assembly outcomes are linked to transcriptions of assembly discussions via LLM tagging, uncovering what, when, who, and where participants deliberate about topics that eventually become proposals/recommendations/outcomes. Furthermore, I analyze the sentiment with which an assembly delegate presented their view on a certain recommendation (agreement, disagreement, etc.) in addition to the supporting reasoning patterns this delegate used to express their view, if any (e.g. whether they draw from personal experience, reference outside expertise, etc.). To evaluate the final prototype tool, I interview subject matter and assembly experts, assembly organizers/facilitators, as well as assembly delegate members to assess the potential and drawbacks of this visualization tool and AI sensemaking backbone. Positive feedback obtained from these user studies include the clear potential for research, narrative building, and facilitation improvement, in addition to greater perceived transparency into the workings of an assembly process. Further work is still needed, however, to address significant lingering issues, such as adjusting presentation to better serve specific use cases and to reduce complexity and confusion, the most referenced drawback of Delibrary. Overall, my thesis aims to \textbf{build transparent insights into the human-led structures of assemblies, enabling relevant stakeholders--from delegates, policy makers, to the general public--to achieve a better understanding of the assembly process and engender legitimacy perception by illustrating that delegates drawn from all walks of life do have meaningful voice in an impactful process}. By helping to promote this understanding and perception of legitimacy of an effective and respectful deliberation process, I strive to ultimately help scaffold healthier democratic decision-making.

Facilitating Creative Learning: Engaging in a Practice of Care

Thu, 01 May 2025 00:00:00 GMT

Facilitating Creative Learning: Engaging in a Practice of Care Presicce, Carmelo Creative learning is shaped not only by tools and activities, but by relationships. This dissertation explores facilitation in creative learning environments as a relational practice centered on care—not as a set of techniques, but as a deeply human way of being with others, a commitment to creating spaces where people feel supported enough to explore, connected enough to share, and valued enough to express themselves. Grounded in constructionist, socioconstructivist, and humanistic pedagogies, the research draws from my multi-year engagement with Learning Creative Learning (LCL)—an online course and global community for educators—and WeScratch, a series of hands-on, collaborative online workshops introducing educators to creative coding. Through qualitative analysis of small-group facilitation during WeScratch workshops, I explore how volunteer facilitators experience and reflect on their practice. Drawing from three case studies, I examine how care takes shape in the situated, relational work of creative learning facilitation. In particular, I identify three interrelated forms of care: epistemic care, which focuses on what and how people learn; affirming care, which supports what learners value and who they are; and convivial care, which attends to how learners feel and relate to one another in a group. After introducing these three forms of care through the work of individual facilitators, I show how epistemic, affirming, and convivial care are deeply interwoven in practice—at times reinforcing one another, at times pulling in different directions. Facilitators must navigate these tensions in the moment, making situated judgments about when to step in, when to hold back, and how to respond to the evolving needs of individuals and groups. By centering care, this research highlights facilitation as deeply human, relational work that sustains the conditions for creative learning, contributing to the broader and evolving discourse on constructionism. It also makes the case for seeing facilitation as an ethical and political practice. In a time when educational discourse is increasingly shaped by ideals of efficiency and optimization—and the world faces rising authoritarianism and dehumanization—choosing to care is not only pedagogically meaningful, but also politically urgent.

Novel Earth Abundant Catalytic Materials for Abatement of Atmospheric Methane Sources, and Evaluation of Agricultural Deployment Environments

Sun, 01 Sep 2024 00:00:00 GMT

Novel Earth Abundant Catalytic Materials for Abatement of Atmospheric Methane Sources, and Evaluation of Agricultural Deployment Environments Brenneis, Rebecca J. Annual global average temperatures in the past year have already exceeded the international target limit of 1.5°C, and the window to prevent that rise from extending is rapidly closing. The high global warming potential (GWP) and short atmospheric residence time (half-life of around 12 years) of methane make it a critical target for action to slow the pace of climate change in this decade. Yet technological solutions for methane abatement are challenged by methane’s inertness, dilute atmospheric concentrations, and diffuse, variable emissions sources. In this thesis, I propose the use of a bio-inspired, earth-abundant, heterogeneous catalysts as a novel tool for atmospheric and emissions-based methane abatement. Copper zeolites were characterized for their ability to convert low levels of methane, continuously, at low temperatures, for moderate durations, and in the presence of a variety of gaseous mixture influents, designed to mimic atmospheric air at standard temperatures and pressures. Catalytic performance was tested under conditions designed to mimic those found at two of the primary sources of low-level, anthropogenic emissions: ventilation air methane (VAM) and industrial dairy. Laboratory synthesized catalysts were shown to completely oxidize methane at concentrations ranging from atmospheric to 1%, covering the range of subflarable levels. Conversion was demonstrated at temperature as low a 270°C, with complete conversion achievable as low as 350°C, in the presence of 20% oxygen. While the presence of water vapor, nitric oxide, and hydrogen sulfide were shown to partially reduce catalytic efficiency, conversion efficiency was restored with increased temperature. The presence of carbon dioxide, alkanes, ammonia and hydrogen, at industrially relevant concentrations, had no effect on catalytic performance. Finally, atmospheric samples were collected at six industrial scale dairy barns across the Midwest and compared with the simulated laboratory conditions. Dairy samples fell within the ranges tested at the bench scale showing no evidence of any impediment to copper zeolite as a potential abatement tool. Methane concentrations at dairies were shown to be on the order of atmospheric to low 10s of ppmv making copper zeolites the only currently identified abatement strategy to address methane emissions at these locations. While it remains to be shown that these zeolites can provide net greenhouse gas benefit in the conditions required, copper zeolites are a strong option on a short list of technologies to address methane at any subflarable concentration, sources of which comprise 80% of global emissions sources, showing great promise as a climate technology breakthrough.

To Co- Is Human: Designing Technologies That Center Human Connection, Co-creativity, and Calm in the Era of AI

Thu, 01 May 2025 00:00:00 GMT

To Co- Is Human: Designing Technologies That Center Human Connection, Co-creativity, and Calm in the Era of AI Dhariwal, Shruti In an era where Artificial Intelligence (AI) systems are increasingly framed as our “companions” and “co-creators,” this dissertation reclaims “co-” as a fundamental marker of shared human experience—using it as a foundation to reimagine and build technologies that consciously center interhuman connection and co-creativity. Central to this work, we’ve developed CoCo (coco.build)—a general-purpose, real-time co-creative learning platform that empowers young people to engage in a wide variety of safe, shared creative experiences with their peers, spanning creative computing, AI education, digital art, writing, and more. Through the platform, we showcase how digital environments can move beyond isolated modes of learning and creating to support multiple ways of being creative together with others—introducing a new paradigm for real-time digital collaboration. We further illuminate how CoCo has been envisioned as a “self-less” social platform that de-emphasizes comparison-based, self-centric metrics (profiles, likes, followers) prevalent in most online systems for youth. We anchor these interconnected ideas in a unifying theme of “Being. Creative. Together.”—reflecting timeless values that have become especially timely in an era when AI tools can further accentuate individualized digital experiences for young people. We supplement the broader design, technical, practical, and pedagogical contributions of this work by sharing insights and feedback from pilots with over 2,000 young people and educators across diverse settings. Ultimately, we see this dissertation as both a contribution and a call—to preserve the human essence of co-, to distinguish it from the useful, powerful, but instrumental AI interactions, and to shape digital environments that nurture young people’s capacity to co-imagine, co-create, co-learn, co-exist, and co-evolve—with and through one another. Note: This work has been co-developed with Manuj Dhariwal. See https://coco.build/thesis for suggested citation and updates on this work.

Interactive Storybooks for Early AI Literacy

Thu, 01 May 2025 00:00:00 GMT

Interactive Storybooks for Early AI Literacy Pu, Isabella As artificial intelligence (AI) becomes increasingly present in children's everyday environments, there is an urgent need for developmentally appropriate tools that help young learners understand and shape these technologies. To be effective, these tools must not only successfully convey complex concepts but also engage children in ways that are meaningful, accessible, and fun. This thesis introduces the Interactive Storybooks for Early AI Literacy, a series of ten interactive storybooks for children ages 6–9 that combine narrative, mini-games, and scaffolded creative AI interactions to teach core AI and robotics concepts. The storybooks follow an overarching narrative featuring a friendly robot, Doodlebot, who must learn creative tasks with the child's help, framing the child as an AI designer and introducing them to the concept of training AI models through the narrative. The storybooks additionally contain interactive games and activities which help keep kids excited and engaged, while providing structured opportunities to experiment with and explore AI creation tools. First, a pilot study was conducted at a community summer camp with four Interactive Storybooks. Children expressed joy and pride in their AI creations, used the characters as emotional anchors for learning, and began to successfully articulate key AI concepts. Four engagement archetypes emerged: the Reader, the Gamer, the Showcaser, and the Social Connector, each representing a distinct way children interacted with the storybooks. However, despite behavioral signs of engagement, many children described the narrative portions as boring and claimed to prefer games. To explore this tension, a home deployment study compared two versions of the system: a "Books" condition with the full narrative and a "Games" condition with only instructional text. Both conditions included the same mini-games and AI interactions. While children in both groups reported similar levels of enjoyment, those in the Books condition showed significantly higher learning gains, greater increases in perceived knowledge and confidence, and stronger connections to the characters. Children in the Books condition also more frequently referenced the narrative when describing AI concepts and demonstrated more creative and iterative behavior during and after gameplay. Overall, these findings suggest that combining storytelling, gameplay, and creative AI interactions is an effective and engaging approach to teaching AI and robotics to young children. Narrative context appears to support concept recall, deepen emotional investment, and promote thoughtful experimentation, even with complex concepts for this age group, like AI and robotics. Based on insights from both studies, this thesis concludes with six design recommendations for creating developmentally appropriate, emotionally resonant AI education tools for early learners using narrative and play.

Decentralized Machine Learning over Fragmented Data

Thu, 01 May 2025 00:00:00 GMT

Decentralized Machine Learning over Fragmented Data Singh, Abhishek The remarkable scaling of data and computation has unlocked unprecedented capabilities in text and image generation, raising the question: Why hasn’t healthcare seen similar breakthroughs? This disparity stems primarily from healthcare data being fragmented across thousands of institutions, each safeguarding patient records in regulatory-compliant silos. The problem is not limited to healthcare but extends to other industries with fragmented data across institutions and individuals. Instead of centralizing various datasets to solve the fragmentation problem, which raises regulatory and ethical concerns, this thesis proposes systems and algorithms to decentralize the machine learning pipeline. Current approaches in this area have centered around Federated Learning (FL), which enables model training over distributed data. However, FL’s dependence on central coordination and inflexibility with heterogeneous systems limit its applicability in healthcare settings. Motivated by these challenges, I explore the following three core themes: 1) Coordination – Today’s coordination algorithms typically rely on static rules or randomized communication, approaches that turn out to be sub-optimal when data heterogeneity is high. I present a new system and a benchmark framework that enables systematic assessment of different coordination algorithms. Next, I propose an adaptive coordination algorithm that leverages historical performance and learning dynamics to improve coordination. 2) Heterogeneity – Data owners can vary significantly in their data distributions, computational resources, and privacy requirements. To address this heterogeneity, I turn the focus from the traditionally protected training phase to securing the critical inference process. Next, I develop techniques for distributed training that adapt to heterogeneous computational capabilities across different agents. 3) Scalability – Enabling scaling in decentralized ML requires addressing three key challenges: parallelization, synchronization, and self-scaling. While parallelization has advanced significantly, the other two remain challenging. I present a framework for offline collaboration through sanitized, synthetic datasets that eliminates constant synchronization needs while preserving privacy. This thesis identifies and addresses some of the bottlenecks along these three core themes through a complementary set of solutions: adaptive coordination, heterogeneity-aware training, and scalable collaboration. Together, these building blocks can enable a practical framework for unlocking data silos across institutions.

Interplay between spatial structure and competition in ecological communities

Thu, 01 May 2025 00:00:00 GMT

Interplay between spatial structure and competition in ecological communities Swartz, Daniel W. Ecology, much like physics, has a long history of theoretical contribution. In this thesis, we take a physics approach to describing ecological communities, searching for simple, emergent features that can generalize beyond specific models of community dynamics. Unifying all of the models we study is an underlying spatial structure, leading to a richer set of possible behaviors than a typical well-mixed model. We first study the case of a metapopulation, a collection of smaller communities linked by dispersal. We find that when the environment is allowed to fluctuate stochastically, new growth laws emerge at the single species level, and high diversity is achieved in the case with many species. We then study the case of pathogen evolution, again in the metapopulation framework. We find that intermediate dispersal can act as a strong driver of pathogen evolution. We also study what happens as a population of microbes expands into unexplored territory, known as a range expansion. We find that a simple model can capture all morphological phases observed in experiments and predict invasion fitness as a function of local and global competitive ability. We also break a standard assumption in microbial ecology, the isotropy of space, and find that a new sector morphology emerges.

The Algorithmic Cookbook of Quantum Science: Quantum and Classical Recipes for Computation

Thu, 01 May 2025 00:00:00 GMT

The Algorithmic Cookbook of Quantum Science: Quantum and Classical Recipes for Computation Martyn, John Michael Since the dawn of science, computation and physics have evolved alongside each other, both driven by a shared quest to solve problems and calculate properties of the natural world. Today, this symbiotic relationship is epitomized in quantum information science, which proposes to use quantum mechanics to solve hard computational problems and develop new paradigms of communication and cryptography. Yet often absent from these developments is a clear, human-interpretable understanding, with many quantum protocols built from inherently quantum concepts (e.g., entanglement, superposition) that defy our classical line of thought and muddle the search for efficient quantum algorithms. Here we show that this search need not be so opaque: simple mathematical tools, namely polynomials and their fundamental theorems, in unison with concepts from classical computing, provide a powerful framework for the design of quantum algorithms. We develop this framework and use it to construct an assortment of quantum algorithms, including methods for quantum simulation, parallel computing, randomized algorithms, and continuousvariable quantum hardware. In illuminating this framework, we find a striking bidirectional flow: just as classical concepts inspire new quantum algorithms, so too can quantum mechanical insights bring about novel methods of classical computing. In this reverse direction, we adopt inherently quantum concepts, such as random compilation and bosonic symmetry, to develop new classical methods, with applications in simulating quantum systems and designing robust neural networks. In aggregate, this thesis provides a compendium of algorithmic techniques for probing quantum systems and solving hard problems, using both quantum and classical tools—an “algorithmic cookbook”—predicated on deep connections between these two domains. The recipes presented here aim to demystify black boxes of quantum information science, and provide a valuable resource for future developments.

The Death of Quasiparticles: Strongly Interacting Gapless Phases with Fermi Surfaces and Fractional Statistics

Thu, 01 May 2025 00:00:00 GMT

The Death of Quasiparticles: Strongly Interacting Gapless Phases with Fermi Surfaces and Fractional Statistics Shi, Zhengyan The emergence of quasiparticles at low temperature provides a powerful organizing principle for many quantum phases of matter, ranging from conventional magnets and superconductors to exotic insulators with topological order. In this thesis, I describe my research in gapless quantum phases in which the framework of quasiparticles breaks down. The main characters are two categories of gapless phases that feature the interplay between strong interactions and two additional ingredients – Fermi surfaces and fractional statistics. Chapter 2 through Chapter 5 focus on strongly interacting metals with Fermi surfaces. The most salient examples are a class of Hertz-Millis models describing the onset of spontaneous symmetry breaking in a metallic environment. At the quantum critical point, gapless order parameter fluctuations destroy quasiparticles living on the Fermi surface, giving rise to a strongly coupled non-Fermi liquid metal. A key result of these chapters is the identification of an infinite-dimensional symmetry that survives in these non-Fermi liquid metals despite the death of quasiparticles. This infinite-dimensional symmetry and its quantum anomaly lead to a series of non-perturbative results on thermodynamics and transport, which are confirmed by perturbative diagrammatic calculations in special examples. Chapter 6 through Chapter 8 explore quantum phases in which anyonic quasiparticles with fractional statistics play an essential role. When parameters in the system are tuned to close the anyon energy gaps, the original anyons lose their coherence and a variety of novel phases emerge. A highlight in this direction is a new mechanism for topological superconductivity in itinerant abelian and non-abelian anyon fluids, which could make contact with experiments on doped fractional quantum anomalous Hall states in the near future.

Particles Inside Particles: The Flow of Energy in Quarks, Gluons, and Jets

Thu, 01 May 2025 00:00:00 GMT

Particles Inside Particles: The Flow of Energy in Quarks, Gluons, and Jets Alipour-fard, Samuel This thesis presents the author’s work in developing probes of the inner structure of jets in high-energy particle collisions. We begin by introducing QCD and the scattering of partons (quarks and gluons), discussing jets as theoretical and experimental proxies for partonic physics, and presenting the partonic cascade model of jet formation and jet substructure. Noting the ubiquitous presence of low-energy pollution in particle collision events, in the forms of hadronization, detector effects, the underlying event (UE), and pileup (PU), we then move towards the modern research area of developing pollution-insensitive probes of jet substructure. Pollution-insensitive features of jet substructure are often accessed theoretically either through jet grooming or energyweighted correlation functions. We present the basics of the modern theory of jet grooming as well as the work of the author in developing the Piranha paradigm for continuous jet grooming, introduced by the author in Ref. [1], and explore the formal and phenomenological benefits of continuous grooming techniques as pollutioninsensitive probes of jet substructure. We introduce the basics of the simplest energy-weighted correlation function – the energy-energy correlator (EEC), which probes angular correlations between particle pairs – and discuss its multi-particle analogues. We focus on the efficient and visually intuitive projected and resolved energy correlators introduced by the author in Ref. [2], which provide computationally-realistic, pollution-insensitive probes of angular many-body correlations in QCD jets. Finally, we exposit the generic theory of energy-weighted observable correlations (EWOCs), introduced by the author in Ref. [3], which utilizes the energy weighting of the EEC to provide pollution-insensitive probes of non-angular correlations within jets.

Solid-state cavity quantum electrodynamics with spin ensembles

Thu, 01 May 2025 00:00:00 GMT

Solid-state cavity quantum electrodynamics with spin ensembles Wang, Hanfeng Quantum sensors have the potential to operate at fundamental physical performance limits. Among various quantum sensing platforms, solid-state spin emitters stand out due to advantageous characteristics such as room-temperature spin polarization and readout, atomic-scale spatial resolution, and extended coherence times. Despite these strengths, traditional optical detection methods exhibit low readout fidelity in solid-state ensembles, severely limiting their achievable sensitivity. This thesis addresses this limitation by coupling a solid-state emitter ensemble to a microwave cavity, forming a cavity quantum electrodynamics system. Our approach eliminates the need for photon collection required by conventional optical readout methods, and the resulting strongly coupled system allows efficient cavity-based probing of the solid-state spin ensemble. By exploiting the hybrid quantum system with cavity quantum electrodynamics, we achieve record-high sensitivity for solid-state quantum sensors, representing a substantial advancement toward achieving fundamental sensing limits.

Expanding the Phase Space of Photons in Matter: From High-Throughput Screening to Atom-by-Atom Engineering

Thu, 01 May 2025 00:00:00 GMT

Expanding the Phase Space of Photons in Matter: From High-Throughput Screening to Atom-by-Atom Engineering Ghorashi, Ali Focusing on the topological band properties of photonic crystals and the plasmonic properties of two-dimensional metals, we seek to answer the question: what is the phase space of photons in matter? For topology, what are the physical parameters that determine whether a given photonic crystal band hosts Dirac points, a non-zero Chern number, or topologically protected corner states? And for plasmons, what are the experimentally addressable ranges of plasmonic dispersions, phase velocities, confinements, and losses? In particular, is it possible to engineer the elusive lossless plasmon? Using high-throughput screening, artificial intelligence, and atom-by-atom engineering through density functional theory, we determine the topological prevalence of photonic bands, propose two systems that evade plasmonic losses through the electron-phonon interaction, and (re)discover general physical laws that govern the geometries of photonic eigenstates.

On the Sample Efficiency of Data-Driven Decision Making

Thu, 01 May 2025 00:00:00 GMT

On the Sample Efficiency of Data-Driven Decision Making Qian, Jian This thesis studies the fundamental problem of decision making under uncertainty through the lens of statistical decision theory. We characterize the minimax risk, which captures the sample efficiency required for effective decision making across three key settings: offline estimation with batch data, online estimation with sequential data, and interactive decision making as exemplified by multi-armed bandits and reinforcement learning. The first part of the thesis develops novel algorithmic and theoretical tools to enhance decision making in these regimes and to bridge the gaps between them. We revisit logistic regression in the offline setting and provide guarantees without restrictive boundedness assumptions. We then propose meta-algorithms that reduce online estimation to offline estimation, enabling any offline estimator to be used effectively in online scenarios. Furthermore, we present general-purpose algorithms for interactive decision making problems by leveraging offline or online estimation techniques. The second part of the thesis introduces a unified approach to understanding the fundamental complexity of interactive decision making. We propose the Decision Making with Structured Observation (DMSO) framework, which encompasses bandits, reinforcement learning, and more general settings. Within this framework, we develop a new complexity measure—the Decision-Estimation Coefficient (DEC)—which captures both upper and lower bounds for minimax regret. DEC extends classical notions such as the modulus of continuity to interactive scenarios by introducing an adaptive variant of Le Cam’s method. Finally, we unify the three classical lower bound techniques—Le Cam’s method, Assouad’s lemma, and Fano’s inequality—through a generalized formulation that also incorporates the DEC, offering a comprehensive understanding of the minimax risk in decision making tasks.

Towards achieving power autonomy in soft-actuated micro aerial robots

Thu, 01 May 2025 00:00:00 GMT

Towards achieving power autonomy in soft-actuated micro aerial robots Ren, Zhijian Micro aerial robots with insect-like flight capabilities hold immense promise for various applications, including environmental monitoring, precision agriculture, and infrastructure inspection in confined spaces. However, realizing power autonomy in these miniature robotic platforms presents significant challenges due to weight constraints, power density limitations, and inefficient actuation at small scales. This dissertation presents three essential improvements towards achieving power autonomy in soft-actuated micro aerial robots. Our robotic platform is driven by a dielectric elastomer actuator (DEA) and generates lift force through flapping wings, a similar mechanism found in flying insects. First, we implemented a dynamic model to optimize the robot components for pairing with an improved DEA to generate a higher lift force. The robot achieved a peak lift-to-weight ratio of 4.3 and demonstrated a 20-second hovering flight with position and attitude errors smaller than 2.5 cm and 2◦ . Second, we fabricated a lightweight high-voltage boost converter that transformed a 7 V DC input into an AC waveform of 600 V and 400 Hz to drive the actuator. This is the first onboard boost converter that can drive the soft-actuated micro aerial robot to take off, and it represents a substantial achievement in miniaturizing power electronics for microrobots. Third, we took inspiration from the natural autorotation of maple seeds in their slow descent. We implemented the first samara-inspired mechanism on micro aerial robots, enhancing lift generation while maintaining in-flight attitude stability without feedback control. The 1.22-gram vehicle can stably take off in 1 second with a total input thrust of 1 gram-force. These accomplishments provide a pathway towards achieving power autonomy and open opportunities for developing agile, robust, and autonomous micro aerial robots for diverse applications.

Low-Energy Electron-Photon Interactions in a Scanning Electron Microscope

Thu, 01 May 2025 00:00:00 GMT

Low-Energy Electron-Photon Interactions in a Scanning Electron Microscope Simonaitis, John The interaction of free-electrons with matter and light is among the most fundamental of processes in nature. From the use of free-electrons for atomic imaging, to their use in the generation of high-intensity, tunable light in synchrotrons, the physics of unconfined electrons has wide application. In recent years, there has been a new focus on the quantum nature of individual electrons in electron microscopes to enable further improvements in these technologies. This work takes advantage of developments in ultrafast optics, electron spectroscopy, quantum optics, and nanofabrication to explore various electron-electron, electron-photon, and electron-material interactions. In this thesis, we construct a low-energy, ultrafast scanning electron microscope, using it to explore quantum coherent interactions between electrons, light, and matter. In Chapter 1, we review the history of free electron experiments and how advances in nanofabrication, low-dimensional materials, and ultrafast optics have opened new opportunities for electron-light interactions to a degree not previously possible. In Chapter 2 we discuss experimental forms of quantum electron microscopy known as interaction-free measurement and electron multi-passing. Chapter 3 details a general theory of electron-photon interactions, including simulations with quantum two-level systems and extended optical nanostructures. In Chapter 4, we design and construct a second microscope with ultrafast triggering, an electron spectrometer with sub-eV resolution, nanostructured interaction regions, and active beam alignment. Chapter 5 explores various experimental results, demonstrating enhanced loss spectroscopy of 2D materials, energy resolution of gold nanoparticle plasmons, as well as spectroscopy of time-tagged cathodoluminescence from optical fibers. Finally, in chapter 6 we discuss future perspectives of this approach, analyzing the impact a heralded electron source would have on electron microscopy and lithography.

Studies of Jet Modification in Heavy Ion Collisions with the CMS Experiment

Thu, 01 May 2025 00:00:00 GMT

Studies of Jet Modification in Heavy Ion Collisions with the CMS Experiment Park, Mary Isabelle In the Compact Muon Solenoid (CMS) detector at the Large Hadron Collider (LHC), lead ions are collided at ultra-relativistic velocities to produce Quark-Gluon Plasma (QGP), a state of matter where quarks and gluons are deconfined and move collectively. Jets are produced in high-momentum transfer parton scatterings prior to and independently of QGP formation, and serve as natural probes of its properties. As the high-energy partons pass through the QGP, they lose energy through medium-induced gluon radiation and elastic scattering, resulting in jets that are modified with respect to the vacuum baseline. In this thesis, jet modification is quantified by measuring the jet production cross section as a function of jet radius in inclusive jets and the jet axis decorrelation in jets recoiling from isolated photons in Lead-Lead (PbPb) and Proton-Proton (pp) collisions. Both measurements indicate that effects of medium-induced jet broadening may be balanced by survivor bias in PbPb collisions, potentially due to differences in the magnitude of quenching of wide versus narrow jets. The results underline the importance of constraining the initial jet kinematics with bosons, which are unmodified by the QGP.

Understanding Drivers of Stratospheric Ozone Change and Fingerprinting its Recovery

Thu, 01 May 2025 00:00:00 GMT

Understanding Drivers of Stratospheric Ozone Change and Fingerprinting its Recovery Wang, Peidong Stratospheric ozone serves as Earth’s natural protective layer, shielding the surface from harmful ultraviolet radiation. The discovery of the Antarctic ozone “hole” in the late 1980s raised significant societal and scientific concern, prompting the rapid regulation of ozonedepleting substances (ODSs) under international treaties. While the signs of ozone recovery have begun, new challenges continue to arise. This thesis investigates three critical factors driving stratospheric ozone changes and influencing the detection of ozone recovery: (1) ODS emissions, (2) chemical chlorine processes, and (3) internal climate variability. With ODS emissions being regulated under the Montreal Protocol and studies now focusing on illicit new production on the order of tens of gigagrams per year, the ocean’s role as both a natural source and sink of ODSs becomes increasingly important. However, these processes have often been overlooked or highly simplified in past ozone assessments. Using a hierarchy of models, from simple box models to global ocean general circulation models, I quantified the ocean’s uptake and release of various ODSs. Chapter 2 examines the ocean’s uptake of chlorofluorocarbons (CFCs), particularly emphasizing its influence on recent illicit CFC emissions estimation. Chapter 3 extends this analysis to include ocean uptake and potential microbial degradation processes, evaluating their effects on emission estimates for various hydrochlorofluorocarbons (HCFCs) and hydrofluorocarbons (HFCs), which are chemical constituents that have been used to replace CFCs. Once these man-made ODSs reach the stratosphere, they are photolyzed to chlorine reservoir species (e.g., HCl and ClONO2), which, through heterogeneous reactions, can transform into reactive chlorine that depletes ozone. While heterogeneous chlorine activation on volcanic ash is well understood, the unprecedented 2020 Australian wildfires raised new questions about chemical processes on smoke particles. This knowledge gap existed because only a few wildfires had injected significant amounts of smoke particles into the stratosphere during the satellite era. Leveraging over 30 years of satellite data, I separated chemical and dynamic processes affecting chlorine reservoir species to quantify chemical chlorine activation across different aerosol types. In Chapter 4, I developed a new approach to quantitatively estimate the onset temperature for chemical chlorine activation after the 2020 Australian wildfire using satellite observations. Chapter 5 applies this method to compare the impact of chemical chlorine activation from two independent wildfire events with that from a series of volcanic eruptions of varying magnitudes. Despite emerging challenges such as illicit emissions and recent wildfires and volcanic eruptions, advancements in observational records, our understanding of ozone chemistry, and computational power have significantly enhanced our ability to quantitatively detect and attribute stratospheric ozone changes. In Chapter 6, I applied a pattern-based “fingerprinting” technique to quantitatively separate the contributions of ODS forcing from other external forcings and internal variabilities in satellite observations. This analysis shows that Antarctic ozone increases cannot be explained by climate internal variability alone, providing strong confidence that ozone recovery is underway, primarily driven by human efforts to reduce ODS emissions.

Light-Induced Collective Interactions in Arrays of Quantum Emitters

Thu, 01 May 2025 00:00:00 GMT

Light-Induced Collective Interactions in Arrays of Quantum Emitters Rubies-Bigorda, Oriol The interaction between light and matter has captivated physicists for centuries, from early studies of vision and refraction in ancient Greece to the development of quantum mechanics and quantum electrodynamics in the past century. While the response of a single quantum emitter to light is well understood, the radiative properties of an ensemble of closely spaced emitters are far more intricate. Coupling to a shared electromagnetic environment induces coherent and dissipative interactions between emitters, giving rise to a collective response that cannot be captured by treating them independently. In the regime of few excitations, the system hosts delocalized subradiant states, that is, coherent superpositions that are largely decoupled from the electromagnetic field and thus decay at suppressed rates. While this weak coupling makes subradiant states attractive for quantum technologies, it also renders them difficult to manipulate. At higher excitation densities, the intrinsic nonlinearity of emitters and the exponential growth of the Hilbert space make theoretical and numerical descriptions of the system and its dynamics increasingly challenging. This thesis explores two fundamental questions: How can subradiant and dark states be selectively accessed and harnessed for practical applications in quantum technologies? And how can interacting ensembles of quantum emitters be efficiently simulated to uncover their many-body physics? The first part of the thesis develops protocols for controlling and addressing dark states in free-space and waveguide-coupled atomic arrays, demonstrating their utility in quantum storage and the deterministic generation of entangled photonic states. Incorporating atomic motion, we further show that collective subradiant states can enhance cooling in dense atomic arrays, offering new avenues for controlling motional dynamics. In the second part, we introduce cumulant expansions of the equations of motion as a powerful tool to analytically and numerically investigate collective decay in the many-body regime. We first examine the collective decay of fully excited atomic arrays in free space, characterizing the onset and scaling of the superradiant burst across different geometries. In collaboration with experiments on ultracold erbium atoms in optical lattices, we provide the first direct observations of many-body collective effects in free-space ordered arrays, including early-time superradiant bursts, late-time subradiant tails, and the emergence of atomic correlations throughout the dynamics. Finally, we theoretically and numerically explore the transient formation of multi-excitation subradiant states, and demonstrate how the existence of multiple dissipation channels suppresses steady-state superradiance in extended arrays.

Hybrid Core Inductors for High Saturation Capability

Thu, 01 May 2025 00:00:00 GMT

Hybrid Core Inductors for High Saturation Capability Yang, Rachel S. Power electronics are critical for any system requiring electricity and often impact the performance of these systems. In many cases, the performance of power electronics is limited by lossy and large inductors that are constrained by the saturation of their magnetic core material. Such saturation-limited inductors are typically found in power electronics applications where the inductor sees large dc current with relatively small ac ripple, such as EMI filters or converters operating in continuous conduction mode. This thesis investigates two types of inductor designs that can achieve higher saturation capability by combining multiple materials in a single core, enabling these designs to achieve greater energy storage or lower loss than conventional single-material cores. The first design combines a permanent magnet with a soft magnetic material (e.g. ferrite) to form a PM hybrid core. This core achieves higher saturation capability by directing PM flux to oppose winding flux in the ferrite. First-order models, design processes, and other theory for the PM hybrid core are developed in this thesis, and different geometries for this core are explored. Additionally, two PM hybrid core prototypes are presented, one using a pot core geometry and one using a modified E core geometry. The PM hybrid pot core prototype achieves 70% more energy storage or 50% of the dc loss versus comparable ferrite prototypes, while the PM hybrid E core prototype achieves 30% more energy storage or a minimum of 52% of the total loss versus comparable ferrite prototypes. The second design pairs a low-frequency, high-saturation material (e.g. steel) with a low-saturation, highfrequency material (e.g. ferrite) to form a steel hybrid core. This core achieves higher saturation capability by directing most of the dc flux to the steel and all of the ac flux to the ferrite, enabling the core to leverage both materials’ advantages while avoiding their detriments. First-order models and design processes for the steel hybrid core are developed in this thesis. An example steel hybrid core design using a pot core is also presented. This design can achieve 220% more energy storage versus a comparable ferrite prototype, and it may achieve lower loss. Its performance, though, is sensitive to manufacturing and assembly imperfections. In this thesis, both the PM hybrid and steel hybrid cores are demonstrated to have great potential in achieving high saturation capability. By leveraging these hybrid cores, inductor designs can achieve greater energy storage density or lower loss and thus enable higher performance power electronics.

Quantum Gas Microscopy of Bosonic Correlations in the Continuum

Thu, 01 May 2025 00:00:00 GMT

Quantum Gas Microscopy of Bosonic Correlations in the Continuum Xiang, Jinggang This thesis details the complete upgrade and renovation of an existing experimental platform into a high-resolution quantum gas microscope for ultracold 87Rb atoms. Quantum gas microscopes enable site-resolved imaging, providing unprecedented access to quantum statistical effects and many-body phenomena. While such instruments are often employed to study physics in optical lattices, we have innovatively adapted our apparatus to investigate bulk system behavior. A major part of this project involved upgrading the scientific apparatus and retrofitting the previous system. We introduced new optical components, including a high-NA objective, and improved the vacuum system for better optical access. Extensive lab renovations, from upgrading the optical table to reorganizing the laser and imaging setups, were carried out to enhance mechanical and thermal stability. Rigorous optical benchmarking confirmed that the objective achieves diffractionlimited imaging, which is critical for resolving single atoms. This capability allowed us to detect density fluctuations at the scale of the thermal de Broglie wavelength in a quasi-two-dimensional gas of 87Rb atoms. In an experiment resembling Hanbury Brown and Twiss interferometry, we measured a 30% enhancement in the second-order correlation function in situ, demonstrating strong bosonic bunching. This outcome underscores the microscope’s precision and the importance of high-resolution imaging in capturing subtle quantum statistical effects. The successful realization of this apparatus demonstrates the utility of quantum gas microscopes in probing bulk systems. With this new platform in place, future studies can explore critical phenomena, many-body correlations, matter-wave emission, and quantum simulations with cold atoms.

Measurement of Cosmic Ray Lithium Isotopes Using the Alpha Magnetic Spectrometer

Thu, 01 May 2025 00:00:00 GMT

Measurement of Cosmic Ray Lithium Isotopes Using the Alpha Magnetic Spectrometer LaVecchia, Gianni The study of cosmic rays and their properties provides insight into the origins of our universe and is a unique lens on the nuclear physics of the cosmos. The identification of cosmic ray isotopes poses a particular challenge, as it requires the measurement of multiple observables to a high degree of accuracy for the deduction of nuclear mass. Using the unique detection capabilities of the Alpha Magnetic Spectrometer (AMS), the isotope fluxes of cosmic ray lithium in the rigidity range of 1.92 to 25 GV are presented. This work is based on 0.97 million ⁶Li and 1.04 million ⁷Li nuclei collected by the AMS over a 12.5 year period, and improves the error and extent of existing measurements by a factor of 10. These results lead to the conclusion that there is no sizable primary component in cosmic ray ⁷Li. The improvements to the AMS velocity measurement establishes the groundwork for future cosmic ray isotope measurements.

Metrics, Muons, Moments, Models, Machine Learning, Measurements, and More: A Manifesto on Collider Physics

Thu, 01 May 2025 00:00:00 GMT

Metrics, Muons, Moments, Models, Machine Learning, Measurements, and More: A Manifesto on Collider Physics Gambhir, Rikab The interface between particle theory and particle experiments is essential to improving our understanding of the Standard Model and looking for new physics beyond it. At this interface lies a complicated web of complex and expensive simulations that cannot fully be trusted, experimental and theoretical uncertainties, overwhelmingly large amounts of data, all while we have yet to find any deviations from the Standard Model. In this thesis, we propose strategies for improving the theory ↔ experiment pipeline at all stages. We first show how modern Machine Learning and statistical techniques can be used to improve the calibration and resolution of particle detectors in a robust way, which can lead to improved measurement precision. We then develop brand new classes of measurable observables based on the principle of infrared-and-collinear-safety, geometry, and machine learning, which come with guarantees about their theoretical calculability and interpretability, in turn motivating measurements at collider experiments. Finally, we then present two complementary approaches to search for new physics: one, in the form of an experimental proposal for a muon beam dump experiment that is viable alongside a full future collider program; and the other, in the form of machine-learning based anomaly detection to search for subtle signals in already-published data.

Practical Algorithms for Modeling Causality to Accelerate Scientific Discovery

Thu, 01 May 2025 00:00:00 GMT

Practical Algorithms for Modeling Causality to Accelerate Scientific Discovery Wu, Menghua Scientific research revolves around the discovery and validation of causal relationships between variables. Machine learning has the potential to increase the efficiency of this process by proposing novel hypotheses from data observations, or by designing experiments that maximize success rate. This thesis addresses these problems through pragmatic approaches, designed to model large systems and incorporate rich domain knowledge. These algorithms are applied to use cases in molecular biology and drug discovery, which highlight their potential to inform efficient experiment design and to automate the analysis of experimental results.

Recycling and Regeneration of Spent Perfusion Media via Ion Concentration Polarization

Thu, 01 May 2025 00:00:00 GMT

Recycling and Regeneration of Spent Perfusion Media via Ion Concentration Polarization Wynne, Eric Michael The widespread adoption of monoclonal antibody therapies is often constrained by their high prices, which can limit accessibility, particularly for patients in low- and middle-income countries. Addressing this economic barrier is crucial to ensure that life-saving treatments can reach all who need them. We present a series of bioprocessing innovations designed to reduce the cost of monoclonal antibody manufacturing and improve global access to these critical therapeutics. The work focuses on developing technologies for media regeneration and recycling, with the goal of reducing the economic and environmental impact of cell culture media in perfusion mammalian cell culture. We demonstrate a microfluidic separation device engineered to selectively remove metabolic waste products—specifically ammonia and lactate—from spent media using ion concentration polarization. Building on this foundation, a scalable millifluidic system was developed to enable higher-throughput waste removal. We characterized the impact of media recycling upon batch and perfusion cell cultures. We devised a nutrient supplementation strategy to create ‘regenerated’ media that minimized any effect on cell growth and productivity compared to fresh media. To support continuous manufacturing, a perfusion culture system incorporating a microfluidic spiral cell retention device and continuous cell bleed was established, and stable performance was maintained over extended durations. A further innovation introduced a multi-stage waste recovery system that increased media regeneration yield to 87.5%. This recovery rate enabled a self-recycling perfusion bioreactor in which 75% of the media feed was regenerated, without significant impact on cell growth, productivity, or product quality. Together, these advances establish a novel biomanufacturing platform that combines electrokinetic waste removal with media regeneration and recycling. The approach is broadly adaptable to mammalian cell culture processes and offers a promising path toward more sustainable, cost-effective, and environmentally responsible production of monoclonal antibodies and other biologics.

Scaling Cooperative Intelligence via Inverse Planning and Probabilistic Programming

Thu, 01 May 2025 00:00:00 GMT

Scaling Cooperative Intelligence via Inverse Planning and Probabilistic Programming Zhi-Xuan, Tan How can we build cooperative machines that model and understand human minds — machines that assist us with our goals, coordinate on plans, infer the intentions behind our words, and even learn our norms and values? This thesis presents a scalable model-based approach to building such systems via inverse planning and probabilistic programming. First, we introduce a probabilistic programming architecture that implements a Bayesian theory of mind. This architecture, Sequential Inverse Plan Search (SIPS), performs online inference of human goals and plans by inverting a Bayesian model of incremental human planning. By combining high-performance symbolic planners with sequential Monte Carlo (SMC) inference, SIPS achieves faster-than-real-time speed, while scaling to hundreds of possible goals, and remaining robust to human mistakes due to boundedly-rational planning. Second, we present Cooperative Language-guided Inverse Plan Search (CLIPS), a system that integrates SIPS with large language models (LLMs) to model communicative cooperation. By using LLMs as likelihood functions within probabilistic programs, CLIPS can infer human goals from ambiguous instructions, then provide uncertainty-aware assistance with much higher levels of reliability than LLMs can on their own. In addition, CLIPS can be used to infer the shared intentions of communicating agents from their actions and words. Third, we show how inverse planning can model the acquisition of social normativity, formalizing norm-guided societal behavior as a norm-augmented stochastic game (NSG). In NSGs, agents assume that society follows a shared set of social norms, and infer these norms from the actions of other agents. By doing so, agents can rapidly learn cooperative social norms using orders of magnitude less data than model-free approaches. Finally, we present advances in probabilistic programming infrastructure that have enabled architectures such as SIPS and CLIPS. Through interfaces for programmable SMC and probabilistic programming with LLMs, developers can readily compose modeling and inference subroutines when designing probabilistically coherent intelligent systems. Together, these innovations demonstrate the feasibility and scalability of rational AI engineering for cooperatively intelligent machines, while illuminating the computational and algorithmic foundations of human cooperative intelligence.

Atomistic Study of Traveling Skyrmions in Multi-Sublattice Magnetic Materials

Thu, 01 May 2025 00:00:00 GMT

Atomistic Study of Traveling Skyrmions in Multi-Sublattice Magnetic Materials Tremsina, Elizaveta A. The development of novel energy-efficient computing hardware is imperative for the reduction of the carbon footprint and for the extension of computing, mobile and wearable device lifespan. Recent advances have been focused on turning to novel material systems, and one such avenue is magnetic thin films. Bits of information can be encoded by magnetic twisted textures called skyrmions, which can be efficiently driven by applying electrical current. Recently, emphasis has been placed on investigating antiferromagnetic and ferrimagnetic skyrmions, as opposed to the single-sublattice ferromagnetic ones studied earlier, due to their potential for more rapid dynamics and magnetic stability. However, there is a pressing need for a thorough and detailed understanding of the intricacies of skyrmion motion, in particular, limiting velocity, optimization of trajectory, controlled mobility and, notably, the observed dynamic distortions of skyrmion profiles. For this reason, experimental studies are simply not enough to provide a complete picture, since the material parameter space for systems hosting skyrmions is quite large. We perform a comprehensive study combining simulation-based as well as analytical approaches, of the spin-orbit torque motion of skyrmions in a wide host of magnetic materials, ranging from crystalline antiferromagnetic to ferrimagnetic, to ferromagnetic. We systematically analyze the relationship between physical distortions of the skyrmion profiles, based on the action of local Thiele forces, and internal elastic tension forces, providing a quantitative and nuanced explanation of these effects. These results expand the understanding of fundamental properties of magnetic skyrmions, as well as their potential use in spintronics applications.

Transformative Lenses: Empowering Learners with New Perspectives Using Generative AI and Augmented Reality

Thu, 01 May 2025 00:00:00 GMT

Transformative Lenses: Empowering Learners with New Perspectives Using Generative AI and Augmented Reality Leong, Joanne Sau Ling Learning is a fundamental human drive that has been shaped by technological advancements over the years. The emergence of generative AI marks a profound shift—its capacity to produce text, images, and video challenges long‐held beliefs about what only humans could create. This shift creates new opportunities for learning, including enabling the design of more customized and personalized learning experiences. Recognizing that learning is deeply influenced by our perceptions of ourselves, others, and our materials and environments, I propose creating transformative lenses powered by generative AI and augmented reality (AR) to adapt what learners perceive, as a means to empower them with new perspectives. I design and implement a set of novel interactive systems and experiences as case studies that address factors including creativity, communication, and motivation. Studying the use of these systems, I gather early evidence that such lenses can help people to overcome their own limiting thoughts and emotions to move towards realizing their full potential. Reflecting on these case studies, I distill key considerations for designing and applying transformative lenses. Finally, I discuss the broader implications of this work at the evolving intersection of generative AI and learning.

Language Models as Opinion Models: Techniques and Applications

Thu, 01 May 2025 00:00:00 GMT

Language Models as Opinion Models: Techniques and Applications Brannon, William Real-time social media platforms now host the news cycle and shape public opinion, while large language models (LLMs) give us new tools to observe and predict those shifts. This dissertation links the new affordances of social media with the predictive power of LLMs to explain -- and forecast -- opinion change. We first quantify the dynamics of news on an influential social platform, then develop LLM-based tools to forecast persuasion and predict heterogeneous treatment effects (HTEs). Study I — Media tempo and tone. Using 518,000 hours of U.S. talk-radio broadcasts and 26.6 million tweets from elite and mass users, we show that Twitter discourse (i) moves faster at both take-off and fade-out stages of a news event and (ii) sustains greater outrage than radio – despite radio’s often explicitly outrage-focused business model. To our knowledge, this is the first large-scale, data-driven comparison between Twitter and traditional media of both outrage levels and the rate of decay of attention to news. Study II — Zero-shot persuasion forecasting. Across a diverse set of 28 randomized experiments, LLM-based methods outperform an ensemble of strong baselines at predicting HTEs and deliver good performance at predicting average treatment effects (ATEs) — all without any experiment-specific fine-tuning. Study III — Transfer and scaling. Fine-tuning LLMs on contemporaneous news coverage boosts HTE (and ATE) prediction performance greatly, to more than 3x baseline performance. A new minibatch-moment-matching (M3) objective lets us train a 400M-parameter model to nearly match the HTE prediction performance of an 8B model at a fraction of the inference cost. Transfer, however, falters out of distribution on held-out experiments and demographic groups, lending support to contextual theories of persuasion. Overall, we (i) quantify how platform affordances shape the tone and tempo of public discourse, (ii) introduce LLM-based methods that make causal experiments more sample-efficient, and (iii) chart the limits of transfer learning for opinion prediction. Our findings provide practical tools for HTE prediction and help researchers anticipate persuasion dynamics in a media landscape shaped by both humans and machines.

Across the Scales of the Nucleus: Understanding Short Range Correlations from Medium Modification to Probe Independence

Thu, 01 May 2025 00:00:00 GMT

Across the Scales of the Nucleus: Understanding Short Range Correlations from Medium Modification to Probe Independence Denniston, Andrew W. The atomic nucleus presents an intricate system due to the non-linear forces described by Quantum Chromodynamics (QCD) that govern its structure. The range of scales involved is remarkable; the most massive nuclei weigh approximately five orders of magnitude more than the quarks that compose them. The nucleus can be analyzed at various levels, from quarks to hadrons to the nucleus as a whole. Short-Range Correlations (SRCs) within the nucleus play a significant role that spans these diverse scales. At the most fundamental level, SRCs influence the interaction between nucleons. The nucleon-nucleon (NN) interaction, arising from QCD, is crucial in determining nuclear properties. SRCs serve as valuable probes for measuring this NN interaction, as the nucleons within SRCs become effectively decoupled from the rest of the nucleus. Multiple experimental techniques, including electron scattering, have been employed to investigate the NN interaction through SRCs. However, our first project demonstrates that inclusive measurements alone are inadequate to constrain this interaction fully. Moving to the scale of the nucleus, SRCs contribute to the high-momentum tail of the nuclear spectral function. While the low-momentum region is characterized by nucleons exhibiting bulk properties, nucleons begin to pair into SRCs at higher momenta. Our research aims to bridge the understanding between the mean-field portion of the nucleus and its high-momentum SRC components. Additionally, SRCs affect the quark structure of protons, as evidenced by the EMC effect, which indicates that quarks behave differently when protons are embedded within a nucleus—an effect referred to as medium modification. This thesis explores the correlation between SRCs and medium modification across various experimental setups. Finally, we seek to establish an interpretation of the nuclear ground-state. Accomplishing this requires demonstrating that our SRC observables are independent of the probe’s scale and scheme. The concluding project of this thesis illustrates how we utilize triple coincidence quasi-elastic scattering across a range of (Q2 ) values to develop a model-dependent framework for understanding SRC distributions within the nucleus’s ground-state wavefunction.

Discovering and Engineering the Computation Underlying Large Intelligent Agents

Thu, 01 May 2025 00:00:00 GMT

Discovering and Engineering the Computation Underlying Large Intelligent Agents Sharma, Pratyusha The richness of language and intelligent behavior has often been attributed to latent compositional structure. Can we build tools for discovering how deep networks learn and represent this latent structure implicitly? And more importantly, can we use this knowledge to improve generalization in largely structure-less general purpose models or refine our understanding of the world they describe? In this dissertation, I present three perspectives to answer these questions. First, I present experimental methods to functionally characterize the space of learnt solutions in LLMs and demonstrate how this understanding can be used to improve their empirical generalization in a gradient free manner, sometimes by as much as 30% points on language understanding benchmarks. Following that, I show how to decipher the structure of another (black box) language-like system, the naturally arising communication system of sperm whales in the wild, discovering for the first time a unique combinatorial communication system. Finally, I apply insights from these results to equip embodied agents with a latent language of thought—hierarchical and compositional—and show how it can enable long-horizon reasoning and planning in these systems. This dissertation ultimately aims to bridge the gap between natural and artificial intelligence, offering new insights into both the fundamental nature of communication in complex biological organisms in the wild and the development of more powerful, and improved AI systems. A key pattern in the discoveries in this thesis has been how simple structures enable complex externalized behaviors in both biological organisms and AI systems.

Volume Mount Devices

Thu, 01 May 2025 00:00:00 GMT

Volume Mount Devices Han, Alan As Moore's Law ends and AI demands increasingly tax our climate and resources, the limitations of two-dimensional electronics integration have become critical bottlenecks. Surface-mount devices (SMDs) remain entrenched in industry practice despite being insufficient for today's computing challenges and sustainability needs. This thesis introduces the volume mount device (VMD), a three-dimensional electronics packaging standard that bypasses the traditional die-to-server stack while offering a scalable, reversible framework inspired by natural ecosystems' circularity. The VMD approach embeds both electrical function and mechanical structure into modular elements that assemble freely in 3D space. Rather than building circuits on planar PCBs, this system constructs functional circuits by linking components into a self-constraining lattice architecture. My current implementation leverages existing supply chains by incorporating SMD components on small tile PCBs, while establishing a pathway toward eventually replacing SMDs at the IC packaging level. I developed a hybrid assembly system combining 3D printing and pick-and-place automation to build multi-layered electronic assemblies efficiently. Where prior work achieved only tens of parts at hundreds of components per hour (CPH), my system demonstrates automated assembly of hundreds of integrated elements at approximately 1000 CPH. I evaluate various geometric configurations, assess performance overhead compared to conventional approaches, and develop cost-effective, self-aligning connector interfaces for reliable joints—creating a foundation for electronics systems that can be assembled, disassembled, and reassembled as needed while improving resilience against supply chain disruptions.

Generative Diffusion Models Towards De Novo Protein Design

Thu, 01 May 2025 00:00:00 GMT

Generative Diffusion Models Towards De Novo Protein Design Yim, Jason De novo protein design aims to generate proteins with desired functions by rationally engineering novel protein structures and sequences. The structure requires modeling continuous 3D coordinates of atoms with rigid biochemical constraints of the polymer chain while the sequence is a series of discrete amino acids that should fold into a plausible structure. Understanding the protein function-structure-sequence relationship necessary for protein design is complex, but deep learning has proven promising to learn the relationship from large protein datasets. This thesis aims to develop deep learning models that generate novel structures and sequences that can be guided towards desired functions. We first describe novel generative models that learn to generate protein structures and sequences by developing diffusion models over general state spaces including Riemannian manifolds and discrete tokens. The resulting methods – FrameDiff, FrameFlow, and MultiFlow – demonstrate the ability of diffusion models to extrapolate beyond the training data to generate novel and diverse protein structures and sequences that pass in silico protein design filters. Next, we apply diffusion models to practical protein design challenges by collaborating with experimental and computational biologists to develop RoseTTAFold Diffusion (RFdiffusion). By combining the structure prediction capabilities of RoseTTAFold and diffusion modeling principles, RFdiffusion can generate functional proteins with in vitro validated properties such as high-affinity binders and symmetric protein assemblies. De novo protein design aims to generate proteins with desired functions by rationally engineering novel protein structures and sequences. The structure requires modeling continuous 3D coordinates of atoms with rigid biochemical constraints of the polymer chain while the sequence is a series of discrete amino acids that should fold into a plausible structure. Understanding the protein function-structure-sequence relationship necessary for protein design is complex, but deep learning has proven promising to learn the relationship from large protein datasets. This thesis aims to develop deep learning models that generate novel structures and sequences that can be guided towards desired functions. We first describe novel generative models that learn to generate protein structures and sequences by developing diffusion models over general state spaces including Riemannian manifolds and discrete tokens. The resulting methods – FrameDiff, FrameFlow, and MultiFlow – demonstrate the ability of diffusion models to extrapolate beyond the training data to generate novel and diverse protein structures and sequences that pass in silico protein design filters. Next, we apply diffusion models to practical protein design challenges by collaborating with experimental and computational biologists to develop RoseTTAFold Diffusion (RFdiffusion). By combining the structure prediction capabilities of RoseTTAFold and diffusion modeling principles, RFdiffusion can generate functional proteins with in vitro validated properties such as high-affinity binders and symmetric protein assemblies.

Learning to Solve Long-Horizon Robot Manipulation Problems

Thu, 01 May 2025 00:00:00 GMT

Learning to Solve Long-Horizon Robot Manipulation Problems Yang, Zhutian If we want mobile robots that perform multi-step tasks in visually diverse and geometrically complex environments, we need them to quickly decide what to do and how to do it. Manipulating multiple objects in environments with movable and articulated obstacles over time requires the robot to satisfy constraints like collision-freeness, reachability, and action feasibility. For problems with large state spaces, continuous action spaces, and long decision horizons, the hybrid constraint satisfaction problems induced by planners become combinatorially difficult to solve. In this thesis, I will discuss strategies for using offline learning to speed up deploymenttime planning, i.e., using a plan feasibility predictor, a subgoal generator, or a compositional joint continuous constraint solver. I will also present strategies for chaining policies learned from demonstrations using conditional inputs, such as key poses and natural language, for generalization in real-world environments. With the resulting efficient long-horizon manipulation planning system, we can solve complex robotic manipulation problems faster at deployment time. It can also be used to generate diverse large-scale whole-body trajectories as part of the data mixture for training robot foundation models in embodied reasoning, planning, and acting.

Building small domain-specific masked language models vs. large generative models for clinical decision support and their effects on users.

Thu, 01 May 2025 00:00:00 GMT

Building small domain-specific masked language models vs. large generative models for clinical decision support and their effects on users. Sergeeva, Elena The frequently adopted definition of knowledge defines it as “justified true belief”. As one may notice this definition presents some issues when applied to AI: it is unclear to which degree it is justified to use “humanizing” vocabulary like “belief” or “justification” when describing the performance of an AI system. Traditional explicit knowledge-representation based AI involves reasoning over symbolic representation of statements standing for such “justified true beliefs” [1], the modern connectionist methodology however replaces explicit reasoning with making a prediction based on a set of computations done over weighted continuous representations of the inputs. The continuous representations learned by such systems remain “black box-like”, where the only elements directly understandable by the human user are the model inputs and outputs. In the first part of this thesis I introduce a set of Masked-Language model transformer based models for a diverse set of medical natural language processing tasks including Named Entity Recognition, Negation Extraction and Relation extraction that perform as well or better than bigger prompt-and-generate transformer-based causal language models. In the second part of the thesis, I discuss the modern “prompt-and-generate” approach to natural language processing where both the inputs and the outputs of the model are word-like elements commonly referred to as “tokens”. I explore the nature of token based representation of the input and look at the way token “meaning” is refined at each layer of the successive transformer computation. With respect to the outputs, I explore how people engage with AI generated sequences of tokens that people perceive as “explained” predictions.

Language-Centric Medical Image Understanding

Thu, 01 May 2025 00:00:00 GMT

Language-Centric Medical Image Understanding Wang, Peiqi This thesis advances medical image understanding by leveraging the multifaceted roles of language: as supervision, prior knowledge, and a medium for communication. We introduce three main contributions: (1) a weakly supervised framework that uses language in clinical reports to guide fine-grained alignment between image regions and textual descriptions, (2) an adaptive debiasing method that uses language prior to improve the robustness of learning algorithms under noisy supervision, and (3) a novel approach for calibrating linguistic expressions of diagnostic certainty, enabling more reliable communication of clinical findings. Together, these methods lead to more accurate, robust, and reliable machine learning systems, ultimately streamlining clinical workflows and improving patient care.

Exploring spin physics with ultracold atoms

Thu, 01 May 2025 00:00:00 GMT

Exploring spin physics with ultracold atoms Lee, Yoo Kyung The dynamics of many interacting spins is an active frontier of research; not only can they explain magnetic phenomena, but they also provide paradigmatic models with deep connections to high-T_c superconductivity, optimization problems, neural networks, and more. Experiments with ultracold alkali atoms in optical lattices have realized spin models with great success. In particular, the isotropic Heisenberg model---the XXX model---was realized more than a decade ago. The ⁷Li apparatus described here was the first to realize a tunable, anisotropic Heisenberg model, also known as the XXZ model. In this thesis, I will describe how the capabilities of this apparatus were harnessed to characterize the spin models we realize, employ them to observe new resonances, and to contribute to studies in spin squeezing and fundamental physics. First, I will discuss how we prepared and observed phantom helix states: eigenstates of the XXZ Hamiltonian. Our understanding of the contact interactions and the phantom helix states enabled us to observe long-predicted lattice-induced resonances, whose effects can be leveraged as another knob to tune the XXZ Hamiltonian. Furthermore, our control over the spin system allowed us to generate spin-squeezed states, a paradigmatic form of entanglement for spin ensembles. This is the first time squeezed states were realized with nearest-neighbor contact interactions in a lattice. Finally, our control over the spin degree of freedom and defects in our state preparation allowed us to create pristine periodic lattices with which to study gedankenexperiments in light scattering.

Probing the Diversity of Fast Radio Bursts with CHIME/FRB

Thu, 01 May 2025 00:00:00 GMT

Probing the Diversity of Fast Radio Bursts with CHIME/FRB Shin, Kaitlyn Fast radio bursts (FRBs) are extremely bright extragalactic radio transients that flash for microseconds to milliseconds at a time, most never to repeat again. Encoded in every observed FRB is information from burst propagation effects, giving us clues about their mysterious origins as well as the environments they traveled through. With inferred all-sky rates of hundreds per day, FRBs have held great interest for those interested in extreme astrophysical processes as well as those interested in cosmological properties of the Universe. The Canadian Hydrogen Intensity Mapping Experiment (CHIME)/FRB project has revolutionized the FRB field with its field-leading discovery rate. With CHIME/FRB, we can start to carry out population-level studies of FRBs to constrain their origins and inform their use as cosmological probes. I present the first population-level studies of CHIME/FRB-observed FRBs using the CHIME/FRB Catalog 1 data release and the injections system to account for observational biases. I discover that CHIME/FRB is likely observationally biased against bursts originating from turbulent local environments, and constrain the energy and distance distributions of FRBs. I also present the Catalog 1 dataset updated with channelized raw voltage (“baseband”) data (“BaseCat1”), for which I played a pivotal role. The CHIME/FRB baseband localization pipeline can localize FRBs to arcminute-precision as long as the signal is bright enough to trigger the saving of offline baseband data. I then discuss two single source-studies enabled by the baseband localization pipeline — one discovering repeaters during phases of unusually heightened burst activity, and one using the burst properties of an unusual FRB to probe the properties of its sightline. In the latter study, I constrain the electron density content of a diffuse filamentary structure on the outskirts of the Virgo Cluster, demonstrating the power of FRBs as probes of diffuse media.

Controlling for the Ionospheric and Baseline-Offset Uncertainties in the CHIME/FRB Outriggers VLBI Network for Milliarcsecond Precision

Thu, 01 May 2025 00:00:00 GMT

Controlling for the Ionospheric and Baseline-Offset Uncertainties in the CHIME/FRB Outriggers VLBI Network for Milliarcsecond Precision Willis, Jacob Fast radio bursts (FRBs) are a novel form of radio transients discovered in 2007. These bright, extragalactic radio signals have an inferred all-sky rate of hundreds of detections per day. The properties of FRBs hold valuable clues about the extreme physical processes driving them while also holding information about the astrophysical plasmas they traverse on their journey to Earth. The Canadian Hydrogen Intensity Mapping Experiment (CHIME)/FRB project has led the field with the hundreds of FRB detections the collaboration has published to date. However, these detections typically have localization regions so large that we cannot identify a single host galaxy, never mind its local environment. To improve upon this, CHIME/FRB has been transformed into a very long baseline interferometry (VLBI) array, drastically increasing the angular resolution of CHIME/FRB from arcminute to sub-arcsecond precision. In this work, I present my contributions to commissioning the CHIME/FRB VLBI Outrigger station located at the Green Bank Observatory (GBO) in West Virginia. This includes measuring and validating GBO's exact position to enable the localization of FRBs to sub-arcsecond precision. For VLBI networks spanning thousands of kilometers, the difference in the local ionospheric environments is significant and leads to errors in the CHIME/FRB Outrigger localizations. I present a thin shell model of the ionosphere to parameterize the local ionospheric environment for each VLBI station. This model may be used to interpolate the error induced by the ionosphere in FRB observations.

Using Z-hadron correlations to probe the medium response in PbPb and pp collisions at √ˢNN = 5.02 TeV

Thu, 01 May 2025 00:00:00 GMT

Using Z-hadron correlations to probe the medium response in PbPb and pp collisions at √ˢNN = 5.02 TeV Chou, Pin-Chun The first measurement of Z-hadron two-particle correlation function are reported in PbPb collisions at √ˢNN = 5.02 TeV, using the PbPb collision data taken in 2018. The integrated luminosity of the PbPb data is 1.67 ±0.03 nb⁻¹ which made the analysis possible for the first time. Collision data with at at least one Z boson with 40

DePUDS: Decentralized Prosocial Urban Development System

Thu, 01 May 2025 00:00:00 GMT

DePUDS: Decentralized Prosocial Urban Development System Zhang, Yan Urban areas face severe socio-economic and environmental challenges like housing crises, inequity, and environmental degradation, often worsened by traditional zoning practices. These are typically rigid, inefficient, outdated, and susceptible to obstruction by narrow special interests (NIMBYism), failing to engage the broader community or adapt to evolving needs. This dissertation proposes the Decentralized Prosocial Urban Development System (DePUDS), a novel governance framework designed to overcome these shortcomings by empowering informed collective consensus and including the often-silent majority. DePUDS integrates decentralized technologies like blockchain and smart contracts with structured economic incentives, facilitated through an accessible user-friendly Decentralized Application (DApp) to encourage broad participation. This system fosters transparent, inclusive, and equitable urban development. Its core mechanism, adaptive incentive-based zoning, dynamically aligns developer profitability with community-endorsed priorities—such as affordable housing, public amenities, and sustainability—providing flexibility absent in traditional zoning. Employing advanced agent-based simulations enhanced by large language models (LLMs), this research rigorously assesses DePUDS's effectiveness across two distinct case studies: Kendall Square in Cambridge, MA (a dynamic innovation hub) and the Inner Richmond District in San Francisco, CA (a culturally rich but housing-constrained neighborhood). Simulation results demonstrate DePUDS significantly aligns development outcomes with community preferences. In Kendall Square, targeted incentives substantially increased affordable housing and public amenities without hindering private investment. In the Inner Richmond, substantial community-driven incentives successfully unlocked constrained development, markedly reducing displacement risks, boosting affordable housing, enhancing amenity access, lowering carbon emissions via density, and preserving local cultural assets. The comparative analysis underscores DePUDS's versatility, showing its potential to enhance growth in active markets and stimulate development in constrained ones. Key policy implications point towards structured DApp-based community participation, adaptive incentive zoning, and dedicated funding. While acknowledging practical implementation hurdles (legal, economic, technological), the findings affirm the feasibility, effectiveness, and transformative potential of decentralized, incentive-driven urban governance. This dissertation offers significant theoretical contributions, practical policy guidelines, and future research pathways to foster more inclusive, sustainable, and resilient urban communities.

Materializing Light: Real-time, Handheld Fabrication of Programmable Structural Color

Thu, 01 May 2025 00:00:00 GMT

Materializing Light: Real-time, Handheld Fabrication of Programmable Structural Color Myers, Paris G. Structural color is nature’s programmable color palette. While pigments and dyes absorb light to produce color, structural color uses nanoscale, light-reflecting structures to appear iridescently colored. We present MorphoChrome, an optical device for real-time, handheld, programmable structural color fabrication. Analogous to painting with light, MorphoChrome creates multicolor, structurally colored designs by exposing a commercially available holographic photopolymer film to user-controlled wavelengths. Within the device, red, green, and blue laser diodes go through an optical prism, combining light and producing mixed color outputs on the film. Additionally, we introduce a resin-based process to adhere and integrate the structurally-colored film with flexible and rigid objects and diverse making processes. In this thesis, we focus on the device optical design and fabrication, color-mixing, color output UI controller, device aperture tips, and holographic photo-polymer film adherence process. We evaluate the available color space and color resolution, and demonstrate creative fabrication applications.

Biophysical specializations supporting efficiency in neural networks

Thu, 01 May 2025 00:00:00 GMT

Biophysical specializations supporting efficiency in neural networks Toloza, Enrique H.S. Neuroscience and artificial intelligence (AI) research have long enjoyed a synergistic relationship. AI has drawn key inspiration from the organization and function of the brain, while our understanding of the biological processes underlying computation has been profoundly enriched by studying the behavior of artificial systems. As breakthroughs in generative AI continue to transform our world, and as the need for more sustainable artificial neural systems becomes more urgent, the neuro-AI feedback loop has never been more important. AI needs ever more powerful and efficient systems, and neuroscience needs further insight into how our brains work. The development of more brain-like AI promises solutions to both of these problems. Unfortunately, this has thus far been stymied by two critical challenges: 1) how do we identify the features that make a system brain-like and 2) how do we incorporate these features into artificial networks in a useful and interpretable way? To address the first of these challenges, I will use the remarkable structural and biophysical diversity of the brain as an introduction into what it means for a system to be “brain-like.” This will lead us to a discussion of dendrites, the tree-like structures implicated at virtually every length scale of neural computation. Dendrites will thereafter act as the focal point for our study of brain-like computation. Specifically, I will trace how relatively simple biophysical features defined at the subcellular level can transform the computational landscape of large networks of neurons. To address the second of these challenges, it is necessary to discuss several enduring problems in computational neuroscience, broken down as chapters in this thesis. In Chapter 2, I will present the development of a new model of single-neuron dynamics that is realistic enough to capture the rich dynamics of dendritic spiking but efficient enough for use in simulations of thousands of neurons, thereby filling a long unmet need in the field. In Chapter 3, I will describe a solution to the general problem of training neural networks with arbitrary differentiable dynamics, thus opening the door for the study of countless biophysical phenomena in the context of networks that can learn to perform computations. In Chapter 4, I will use these tools to test several longstanding hypotheses regarding the utility of different biophysical features in neurons, performing first-of-their-kind fair comparisons of the computational performance of spiking networks, rate-based networks, and networks with nonlinear and linear dendrites. Finally, in Chapter 5, I will use insights gained from studying dendrites at the network level to provide a new perspective as to how the structural and biophysical diversity of the brain could emerge from a complex interplay of functional pressures (e.g., task demands) and physical constraints (e.g., space and energy). Together, the chapters of this thesis outline a general quantitative framework for building more brain-like AI for use in both AI research and neuroscience. This framework illustrates how biophysical specializations arising at the level of single neurons shape the emergent dynamics of the brain.

Next Week Tonight: Simulating Counterfactual Narratives of the future using Agentic Knowledge Graphs

Thu, 01 May 2025 00:00:00 GMT

Next Week Tonight: Simulating Counterfactual Narratives of the future using Agentic Knowledge Graphs Agarwal, Gauri Understanding the ripple effects of events—both real and speculative—is essential for navigating complex futures. Large Language Models (LLMs) have emerged as powerful tools that offer a user-friendly and narrative experience for question answering and reasoning across large corpuses of unstructured data [15, 96]. While LLMs can respond to complex ‘what-if’ questions, they typically provide single, unverifiable answers. Even with retrievalaugmented generation (RAG) that grounds LLM responses on external sources, the opacity of reasoning pathways undermines trust in model outputs [97]. Next Week Tonight builds on the narrative and reasoning capability of LLMs further by enhancing the exploration of what-if futures and making it more transparent and evidencebased. NWT exposes the underlying knowledge graph, allowing users to inspect inference pathways directly. This also enables the generation of multiple, diverse scenarios from a single condition—each following different but explainable causal chains. In testing 15 counterfactual prompts that span diverse news topics, NWT produced scenario narratives that were rated as significantly more causally coherent, transparent, and easier to audit than standard chat completions. Beyond technical performance, NWT reinvents scenario planning as an interactive narrative experience - encouraging curiosity, critical thinking, and deeper engagement with the complexities of future events. By surfacing not only what could happen but why and how, NWT aims to empower analysts, policymakers, and the public to navigate uncertainty with greater clarity and confidence. Github: https://github.com/viral-medialab/next-week-tonight

On the Learnability of General Reinforcement-Learning Objectives

Thu, 01 May 2025 00:00:00 GMT

On the Learnability of General Reinforcement-Learning Objectives Yang, Cambridge Reinforcement learning enables agents to learn decision-making policies in unknown environments to achieve specified objectives. Traditionally, these objectives are expressed through reward functions, enabling well-established guarantees on learning near-optimal policies with a high probability — a property known as probably approximately correct (PAC) -learnability. However, reward functions often serve as imperfect surrogates for true objectives, leading to reward hacking and undermining these guarantees. This thesis formalizes the specification and learnability of general reinforcement-learning objectives beyond rewards, addressing fundamental questions of expressivity and policy learnability. I examine three increasingly expressive classes of objectives: (1) Linear Temporal Logic (LTL) objectives, which extend conventional scalar rewards to temporal specifications of behavior and have garnered recent attention, (2) Computable objectives, encompassing a broad class of structured, algorithmically definable objectives and (3) Non-computable objectives, representing general objectives beyond the computable class. For LTL objectives, I prove that only finitary LTL objectives are PAC-learnable, while infinite-horizon LTL objectives are inherently intractable under the PAC-MDP framework. Extending this result, I establish a general criterion: an objective is PAC-learnable if it is continuous and computable. This criterion facilitates the establishment of PAC-learnability for various existing classes of objectives with unknown PAC-learnability and informs the design of new, learnable objective specifications. Finally, for non-computable objectives, I introduce limit PAC-learnability, a practical relaxation where a sequence of computable, PAC-learnable objectives approximates a non-computable objective. I formalize a universal representation of non-computable objectives using nested limits of computable functions and provide sufficient conditions under which limit PAC-learnability holds. By establishing a theoretical foundation for general RL objectives, this thesis advances our understanding of which objectives are learnable, how they can be specified, and how agents can effectively learn policies to optimize them. These results contribute to the broader goal of designing intelligent agents that align with expressive, formally defined objectives—moving beyond the limitations of reward-based surrogates.

Solid-State Quantum Memories for Near-Term Quantum Repeaters

Thu, 01 May 2025 00:00:00 GMT

Solid-State Quantum Memories for Near-Term Quantum Repeaters Sutula, Madison M. Over the past decade, quantum computers have emerged as a promising technology to enable transformational advances in information processing and communication and solve problems that are intractable to classical computers. While there is great promise in linking quantum computers together over long distances via quantum channels, these technologies are still under development. Solid-state emitters with coherent spin-photon interfaces, long spin lifetimes, and narrow optical transitions are a leading platform for use as quantum memories in networked quantum repeaters. However, while such emitters have already enabled advanced quantum networking demonstrations in laboratory settings, applying these devices as useful memory devices at scale is a key outstanding challenge. In this thesis, we experimentally investigate solid-state quantum memories for quantum information applications. First, we develop experimental techniques to characterize solid-state emitters with high throughput, enabling both better understanding of the distribution of emitter properties and improved feedback on material preparation and device fabrication. Next, we implement quantum frequency conversion to create a coherent spin-photon interface between silicon-vacancy centers in diamond and optical photons in the low-loss telecom band. Finally, we investigate color centers in other engineering materials, including silicon and silicon carbide, to better understand the fundamental trade space of requirements for solid-state hosts. Together, these efforts represent a significant advance in creating, controlling, and deploying telecom-compatible spin interfaces, paving the way for memory-enabled quantum repeaters.

Inferring Clonal Dynamics in Blood using Single-Cell Measurements

Thu, 01 May 2025 00:00:00 GMT

Inferring Clonal Dynamics in Blood using Single-Cell Measurements Perry, Andrea N. In this work, we uniquely tag hematopoietic (blood) stem cells with genetic barcodes and follow their progeny over time to ask whether clonally related cells in myeloproliferative neoplasms (MPNs) favor particular blood cell fates. Myeloproliferative neoplasms are clonal disorders driven most frequently by the JAK2-V617F mutation, which arises in a single hematopoietic stem cell (HSC) and ultimately dominates the normal process of blood cell production. Although all patients carry the same driver mutation, they still branch into three distinct disease forms—essential thrombocythemia (ET), polycythemia vera (PV), or primary myelofibrosis (PMF)—and the reason for this variation remains unknown. One compelling hypothesis is that the JAK2-V617F mutation may arise in HSC subsets with intrinsic biases toward platlet-producing cells (as in ET) or red blood cell precursors (PV). To investigate this question, we analyzed bone-marrow cKit⁺ cells from mice engineered for inducible MPN disease and CRISPR array repair lineage tracing (CARLIN), using single-cell RNA sequencing. Our gene expression analysis shows that the mutation keeps key signaling and stress-response genes switched on and boosts growth-promoting enzymes, collectively pushing blood production toward the myeloid line. At the resolution of individual CARLIN clones (i.e. cells grouped by a shared progenitor), however, we observe no robust mutation-induced lineage bias—an outcome attributable to limited clone recovery and inter-mouse variability. Crucially, this work establishes a scalable analysis pipeline for future, higher-yield CARLIN experiments. Enhancing lineage-tracing sensitivity, barcode diversity, and biological replication will be essential to test whether these interferon-/stress-response and kinase programs manifest as subtle, clone-level fate biases in JAK2-driven MPN.

Principled Approaches for Latency Reduction in Networking Systems

Thu, 01 May 2025 00:00:00 GMT

Principled Approaches for Latency Reduction in Networking Systems Pit-Claudel, Benoit Modern networks face unprecedented challenges due to exponential growth in traffic demands, driven by AI workloads in datacenters and the ubiquitous adoption of cloud services across the internet. This dissertation addresses three critical challenges in network systems: efficient scheduling of inference tasks, performance optimization in hybrid networks, and memory-efficient load balancing in datacenters. First, we introduce Nona, a stochastic scheduling framework that leverages queueing theory to optimize task placement in datacenter environments. By employing randomized algorithms and considering both network and compute constraints, Nona demonstrates multiple orders of magnitude improvements in job completion times while maintaining implementation simplicity. Nona proposes stochastic scheduling, in which the complexity of the scheduling problem is moved to an offline phase. When handling jobs online, stochastic schedulers are oblivious to the instantaneous state of the network and only rely on predetermined allocation probabilities to make lightning-fast decisions. Second, we present LINC, an in-network coding solution designed for hybrid backbone networks. Through comprehensive mathematical analysis and simulation, we highlight the benefits of network coding in cases where no modifications of the end-hosts are possible. Finally, we develop Sirona, a memory-efficient version of a reactive subflow spraying mechanism suited for hardware deployment. We show that Sirona can achieve competitive performance in homogeneous and heterogeneous datacenter networks while keeping a low memory footprint.

Forward Modeling for Bolometry and Disruption Mitigation in Tokamaks or How to Kill Your Plasma With Confidence, Style, and Pizzazz

Thu, 01 May 2025 00:00:00 GMT

Forward Modeling for Bolometry and Disruption Mitigation in Tokamaks or How to Kill Your Plasma With Confidence, Style, and Pizzazz Stein-Lubrano, Benjamin The tokamak is a promising approach to magnetic confinement fusion. Tokamak functionality is threatened by plasma disruption events, which can damage critical machine components. Disruption damage can be mitigated by high-Z impurities, delivered by Massive Gas Injection (MGI) or Shattered Pellet Injection (SPI). Impurities radiate energy out of the plasma and onto the first wall. Evenly distributed radiation causes less damage than unmitigated disruption pathways, which deliver concentrated heat loads. In order to successfully develop and deploy mitigation systems, it is important to accurately measure and characterize disruption radiation. Accurate measurement is challenged by fast disruption timescales and highly asymmetric radiation patterns, which push the time and spatial resolution limits of radiant heat sensors. Previous radiation analysis approaches are typically limited to two dimensions or less by the highly under-determined nature of tomographic reconstruction and limited spatial resolution of sensors. Two dimensional analysis is often inaccurate for disruption radiation, which can be highly three dimensional as a result of localized impurity sources and fast 3D MHD events. In this thesis, I present a new algorithm for 3D radiation analysis in tokamak disruptions, called Emis3D. When Emis3D is applied to mitigated disruptions on the JET tokamak, a significant injection plume radiation effect in mitigated disruptions is revealed. When this effect is included in radiated energy calculations, the mitigated radiation fraction of plasmas with high thermal energy content is significantly improved, indicating that thermal mitigation is more effective than previously thought. Emis3D can also be used as a design tool to evaluate potential radiant heat sensor layouts. When applied to the SPARC tokamak, Emis3D demonstrates that toroidally skewed sensor sightlines improve spatial resolution and reduce blind spots, allowing more accurate measurement.

Understanding the Milky Way with Stars

Thu, 01 May 2025 00:00:00 GMT

Understanding the Milky Way with Stars Ou, Xiaowei "How do galaxies form?" is one of the most important questions in modern astrophysics. Hierarchical growth, the most plausible theory behind galaxy formation, suggests that galaxies, including the Milky Way, assemble through the accretion of smaller systems, over a scaffolding of the invisible Dark Matter. Such growth is evidenced by the differences in stellar structures found in the Galaxy over the last few decades, accelerated most recently by the Gaia space mission. Yet, we still lack a full picture of the formation of the Milky Way and its stellar components, and we are even further in understanding its underlying Dark Matter distribution. For the latter, discrepancies between observations and predictions from CDM model at galactic scales have sparked debate about how well this model accounts for the evolution of the Milky Way. Stellar tracers provide a powerful tool for examining these discrepancies, helping us explore the hierarchical assembly of galaxies in the Local Group and test different models for dark matter. At the same time, cosmological simulations and machine learning techniques offer a bridge between the theory and observations. In this thesis, I combine observation of stellar kinematics and chemistry with cosmological simulations to understand the formation and evolution of the Milky Way and its satellite dwarf galaxies. I map the dark matter distributions in the Milky Way and one of its ultra-faint dwarf galaxies using stellar dynamics, combining simulations of tidal disruption with observational data to study ongoing merger events and how hierarchical assembly shaped the Milky Way today. I conduct robust machine learning searches of kinematic substructures from disrupted dwarf galaxy debris in the Milky Way and utilize stellar heavy element abundances to probe the galaxies that merged with the Milky Way in the past. Lastly, I develop synthetic surveys from simulations to bridge gaps between theory and observation, testing the robustness of current and future methodologies in understanding how the Milky Way came to be.

Coating Thermal Noise in Gravitational-Wave Detectors

Thu, 01 May 2025 00:00:00 GMT

Coating Thermal Noise in Gravitational-Wave Detectors Demos, Nicholas The direct detection of gravitational waves, originating from cataclysmic events such as black hole and neutron star mergers, has ushered in a new era of observational astronomy. These signals offer unique insights into astrophysical phenomena and fundamental physics, but fully realizing their potential requires continued improvements in detector sensitivity. A primary factor limiting the performance of current ground-based interferometers like Advanced LIGO and Advanced Virgo is thermal noise arising from the highly reflective multilayer coatings on the test mass mirrors. Reducing this coating thermal noise, particularly its Brownian component, while simultaneously maintaining exceptionally low optical absorption and scatter is necessary to advance detector capabilities. This thesis addresses this challenge through the characterization and development of alternative coating materials and designs. Central to this work is a dedicated experimental apparatus employing a high-finesse folded optical cavity and a multimode co-resonance technique. This system enables direct, high-precision measurements of coating thermal noise in the frequency band relevant to gravitational-wave detectors and allows for relatively rapid evaluation of candidate coatings, providing timely feedback for materials development. Coating materials such as niobia-based oxides, hafnia-tantala mixtures, and substoichiometric silica, were explored employing strategies like compositional optimization, post-deposition annealing, and multimaterial designs with buried layers. Progress toward lower-noise coatings is demonstrated. Highly reflective coatings based on optimized titania-silica, titania-germania, and ternary silicon nitride structures achieved thermal noise levels approximately 75% that of current detector coatings. These coatings also exhibited exceptionally low optical absorption, reaching levels near 1 part-per-million following appropriate heat treatment. While challenges related to defect formation during annealing and discrepancies between different noise measurement methodologies were identified, ongoing research, particularly on defect mitigation in materials like titania-germania, continues to advance the field. The findings presented here contribute to the materials science foundation for improving current gravitational-wave detectors and guiding the design of future observatories.

Time-Domain Astrophysics with the Transiting Exoplanet Survey Satellite

Thu, 01 May 2025 00:00:00 GMT

Time-Domain Astrophysics with the Transiting Exoplanet Survey Satellite Jayaraman, Rahul The Transiting Exoplanet Survey Satellite (TESS) is conducting an all-sky survey with the primary aim of detecting planets orbiting nearby stars. However, its large field of view and 200 s imaging cadence are useful for other science cases, ranging from stellar astrophysics to transient science. This thesis focuses on using TESS to study both the circumstellar environment and stellar interiors, as well as using the satellite to detect and characterize optical emission from gamma-ray bursts (GRBs). Chapter 2 focuses on the discovery of HD 135348, a "rigidly rotating magnetospheric" star–wherein the stellar magnetic field traps dust in a co-rotating orbit and leads to complex periodic photometric modulations–using solely photometric data. Chapter 3 focuses on the discovery of a long-period subdwarf B (sdB) star using 20 s cadence TESS data and proposes a novel formation mechanism for long-period sdB stars that relies upon stable, nonconservative mass transfer. Chapters 4 and 5 focus on pulsating stars in close binaries, and the evolutionary insights that these "tidally tilted" pulsations enable. In particular, we focus on developing models to track the amplitude and phase of these pulsations as a function of orbital phase, as well as tools to perform physically-motivated modeling of the binary components. Chapters 6-7 focus on the optical signatures of gamma-ray bursts in TESS, and analyze the prompt optical flash that is often observed contemporaneously with the high-energy emission from these bursts. Chapter 7, in particular, aims to connect the prompt optical flash to the high-energy spectral energy distribution (SED), and explains the suppression of the optical flash (compared to the extrapolation of the high-energy SED) by invoking dust extinction in the host galaxy. This thesis represents a significant step forward in both stellar and transient astrophysics; throughout this work, we emphasize the use of an unconventional tool–TESS–to pursue timely scientific questions.

Building Intelligence that can Interact with the Physical World

Thu, 01 May 2025 00:00:00 GMT

Building Intelligence that can Interact with the Physical World Wang, Tsun-Hsuan (Johnson) Recent advances in Artificial Intelligence (AI) have demonstrated remarkable success in parsing, reasoning, and generating digital content across modalities such as natural language, speech, images, videos, and 3D data. However, these breakthroughs have yet to extend meaningfully beyond the digital realm into the physical world. Developing AI for physical interaction poses challenges such as limited grounding, scarce physical data, and high reliability demands in safety-critical settings. This thesis takes a holistic approach to building intelligence that can interact with the physical world – through the lenses of data, brain, and body. Data is the fuel powering highly capable AI systems. We present methods for data-driven simulation that synthesize sensor measurements from physical processes, and knowledge-driven simulation that leverages large language models to generate actor behaviors and scenarios. By reverse engineering the generative processes behind physical data, we address data scarcity while enabling scalable and effective evaluation. The brain, driven by data, demands a deep understanding of the physical world and reliable interaction with it. We introduce methods to bridge the internet-scale knowledge of digital AI with the physical world to improve generalization and interpretability. For greater reliability, we integrate control-theoretic modules into AI models to enable certifiability. Beyond the behavioral intelligence, the body plays a crucial role in physical interaction. We demonstrate how morphological intelligence can emerge from computation and show how pre-trained generative AI models (brain), when augmented with physics-based simulation that provides feedback on generated data, can be applied to robot design. To this end, this thesis explores how digital AI can be extended into the physical world through a comprehensive investigation of data, brain, and body – laying the groundwork for building physical AI.

Biomolecular Modeling at Scale

Thu, 01 May 2025 00:00:00 GMT

Biomolecular Modeling at Scale Wohlwend, Jeremy Predicting the structure and interactions of biomolecules is a fundamental problem in computational biology, with broad implications for disease understanding and drug discovery. Advances in deep learning have enabled remarkable progress, but scaling these approaches to the varied and complex realities of biology is a persistent challenge. This work introduces deep learning methods for biomolecular modeling at scale, designed for efficiency, adaptability, and accessibility. The early chapters present models developed in the general molecular domain, including prediction of structure and interactions for proteins, nucleic acids, and small molecules. To demonstrate how these methods extend to specific biological problems, the latter portion of this work focuses on modeling T cell receptor recognition. As a key immunological mechanism, it highlights the promise of scalable models, but also their present limitations in capturing fine-grained molecular selectivity. Together, these contributions define a framework for bridging foundational models and domain-specific applications, with the potential to scale, and meet the demands of increasingly complex biological systems.

Multimessenger signatures of compact binaries

Thu, 01 May 2025 00:00:00 GMT

Multimessenger signatures of compact binaries Mo, Geoffrey Kwan Lok Gravitational waves and electromagnetic observations provide complementary views into some of the most extreme objects in the Universe. In this thesis, I present studies of multimessenger compact binaries from two angles: electromagnetic follow-up of gravitational-waves, and gravitational-wave follow-up of electromagnetic sources. I first describe technical and computational efforts to enable the distribution of alerts of kHz gravitational-wave sources as a member of the LIGO--Virgo--KAGRA collaboration, and to improve localizations of these events by folding in galaxy catalog information. I then detail work to enable electromagnetic follow-up observations of binary neutron star and neutron star--black hole mergers with two telescopes, the Transiting Exoplanet Survey Satellite (TESS) and the Wide-field Infrared Transient Explorer (WINTER). Approaching multimessenger observations from the opposite direction, I describe a search for gravitational waves coincident with fast radio bursts from the only Galactic fast radio burst source. Lastly, I perform an electromagnetic study of Type Ia supernovae in the mid-infrared, whose white dwarf binary progenitors will be mHz gravitational-wave sources for the future LISA space mission.

Efficient Network Systems Design for Machine Learning

Thu, 01 May 2025 00:00:00 GMT

Efficient Network Systems Design for Machine Learning Yang, Mingran Machine learning (ML) is transforming modern life by powering a diverse range of groundbreaking applications. As ML models and datasets expand, the scale of training and inference workloads in modern datacenters is increasing at an unprecedented pace. As the demand for computing resources grows, the need for low-latency and energy-efficient network systems becomes increasingly urgent. This thesis introduces efficient network systems designed to support machine learning workloads. It presents three key systems: Trio-ML, which accelerates ML training; Lightning, which enhances ML inference efficiency; and on-fiber photonic computing, a forward-looking vision for next-generation computing systems. The first system, Trio-ML, accelerates data-parallel distributed ML training by leveraging in-network computing on Juniper Networks' programmable chipset Trio. Trio-ML features two key designs: in-network aggregation, which utilizes Trio packet processing threads to aggregate gradients directly inside the network, and in-network straggler mitigation, which utilizes Trio timer threads to detect and address stragglers. We prototype Trio-ML on a testbed with three real DNN models (ResNet50, DenseNet161, and VGG11) to demonstrate its effectiveness in mitigating stragglers while performing in-network aggregation. Our evaluations show that when stragglers occur in the cluster, Trio-ML outperforms today's state-of-the-art in-network aggregation solutions by up to 1.8x. The second system, Lightning, is the first reconfigurable photonic-electronic smartNIC to serve real-time ML inference requests. Lightning uses a fast datapath to feed traffic from the NIC into the photonic domain without creating digital packet processing and data movement bottlenecks. To do so, Lightning leverages a novel reconfigurable count-action abstraction that keeps track of the required computation operations of each inference packet. Our count-action abstraction decouples the compute control plane from the data plane by counting the number of operations in each task and triggers the execution of the next task(s) without interrupting the dataflow. We evaluate Lightning's performance using four platforms: prototype, chip synthesis, emulations, and simulations. Our simulations with large DNN models show that compared to Nvidia A100 GPU, A100X DPU, and Brainwave smartNIC, Lightning accelerates the average inference serve time by 337x, 329x, and 42x, while consuming 352x, 419x, and 54x less energy, respectively. Building on the in-network computing and photonic computing concepts discussed in Trio-ML and Lightning, we present a forward-looking vision for future computing systems. We argue that pluggable transponders are a prime platform for performing photonic computing inside the network without having to replace networking switches and routers. Optical transponders are ubiquitous in today's wide-area and datacenter networks, giving us a unique opportunity to re-purpose them for photonic computing. To this end, we introduce on-fiber photonic computing, explore key research challenges in bringing this vision to reality, and discuss real-world applications.

Wireless, Battery-Free, High-Sensitivity 5G RF Energy Harvesters for Next Generation IoT Sensor Tags

Thu, 01 May 2025 00:00:00 GMT

Wireless, Battery-Free, High-Sensitivity 5G RF Energy Harvesters for Next Generation IoT Sensor Tags Yildirim, Deniz Umut The Internet of Things (IoT) is revolutionizing various industries, enabling a new wave of smart applications such as automated asset tracking in warehouses, substation monitoring in smart grids, and precision agriculture. However, as IoT devices proliferate, powering these devices in a sustainable and maintenance-free manner has become a critical challenge. Traditional IoT systems rely on batteries, which present issues of limited lifespan, environmental impact, and maintenance costs, especially in large-scale deployments. As a result, the development of battery-free IoT devices powered by ambient energy harvesting has gained significant attention. Among various energy-harvesting technologies, radio frequency (RF) energy harvesting has emerged as a promising solution for powering IoT devices. By harvesting energy from ambient RF signals in licensed frequency bands, RF energy-harvesting systems eliminate the need for batteries and allow for continuous, maintenance-free operation. This is especially crucial in environments where battery replacement is impractical or impossible, such as in large industrial warehouses, remote infrastructure, and hazardous environments. However, achieving high sensitivity and reliable operation in RF energy-harvesting systems poses several challenges. High-sensitivity rectifiers are required to capture and convert weak RF signals into usable energy, but integrating these rectifiers with ultra-low power baseband data processing circuits remains a significant hurdle. Moreover, antenna-rectifier matching calibration must be compatible with the duty-cycled operation of these tags, where brief communication periods are followed by long charging intervals. Additionally, the antenna system must be robust to detuning when placed on various objects, ensuring that the system can operate effectively in diverse environments. This thesis presents two integrated circuits to work towards these goals. The first chip is designed with the goal of minimizing the charging time as much as possible, which is critical in scenarios such as inventory management in warehouses, and tamper detection. The goal was to achieve < 1-minute charging time while maintaining sensitivity competitive with the state-of-the-art. Unlike previous harvesters that either focused solely on sensitivity without integrating baseband processing and communication, or included those features but considered continuous communication at low sensitivity, the IC developed in this work achieves a sensitivity of −31 dBm and is capable of backscattering data approximately 18 seconds after a cold start. It also provides a detailed description of the difficulty of achieving higher sensitivities at higher 5G frequencies. The second chip in this thesis builds upon the first one and integrates an analog front-end to convert sensor data for environmental monitoring. We implemented an antenna-rectifier calibration method that is maintained as long as there if RF power, even though the tag goes into long charging periods. Even though the charging time, or the data readout interval, for these tags is more relaxed compared to the inventory management applications, we have also developed a design methodology to minimize the energy required to generate a data packet for backscattering, through which we were able to keep the charging time at 4 minutes while having additional functionalities and backscattering at a higher data rate compared to the first chip. Finally, a simple shielding method was implemented to enable the tags to be placed on any objects without resonance frequency detuning. All of these were achieved while still obtaining a sensitivity of −30 dBm, competitive with the state of the art. In addition, the third project investigates the use of heterogeneously integrated “beyondCMOS” devices to enhance overall rectifier performance. These emerging devices, fabricated by Palacios Group at MIT, show promise in overcoming sensitivity limitations commonly found in rectifiers, thereby extending the range and coverage of energy-harvesting IoT systems. We conduct a detailed characterization of these devices, highlighting their unique physical behaviors not present in standard CMOS technology, and provide system-level design guidelines for building improved rectifiers. Preliminary simulation results show that rectifiers using negative-capacitance field-effect transistors (NCFETs) can harvest up to four times as much power than their CMOS-based counterparts, while maintaining the same sensitivity. This thesis outlines the design, implementation, and evaluation of all three systems. The two aforementioned ICs are tested both in simulation and in real-world scenarios such as a typical office environment. Meanwhile, the novel device technologies are explored through simulation, demonstrating their significant potential for next-generation rectifier design.

Additive Manufacturing of Electrical Machines and Electronic Devices

Thu, 01 May 2025 00:00:00 GMT

Additive Manufacturing of Electrical Machines and Electronic Devices Cañada Pérez-Sala, Jorge Recent advancements in the additive manufacture of electronics and electrical machines have led to successful demonstrations of 3D-printed passive (e.g., resistors, capacitors, inductors) and active (e.g., transistors) electronic components, as well as magnetic cores and power transfer devices. However, each new demonstration of 3D-printed functional devices has typically required increasingly specialized and expensive manufacturing hardware. This work opposes that trend by developing a technology capable of fabricating all such devices on a single, affordable machine: a material extrusion 3D printer. Material extrusion stands out among additive manufacturing technologies for its widespread availability and its compatibility with monolithic multi-material manufacturing, essential for the fabrication of functional electromagnetic devices. These attributes, together with its well-established ability to fabricate mechanically functional parts, make material extrusion a promising technology for the single-step fabrication of electronics and electrical machines, and for their monolithic integration into complex devices, such as custom functionalized prostheses, robots, and space exploration hardware. In this research, a desktop 3D printer was transformed into an almost-universal manufacturing machine capable of fabricating a myriad of electrically, magnetically, and mechanically functional devices, using various feedstock formats (e.g., filament, pellets, ink). With this machine, milestones such as the fabrication of the first semiconductorfree, fully 3D-printed logic gates, and that of the first fully 3D-printed motor, have been achieved. Built for under $4000 in parts, the modified 3D printer opens the door to the democratization of electronics and electrical machine manufacturing, empowering institutions and individuals alike, and serving as an educational tool to introduce advanced manufacturing to new generations. Additionally, this work investigates optimization strategies for planar inductors and alternative techniques for the creation of miniaturized, three-dimensional, electrically functional components via two-photon polymerization. By demonstrating novel methods and applications, this thesis advances the state of the art in the additive manufacture of electromagnetic devices and paves the way toward the decentralized fabrication of electrical machines and electronic devices.

A Unified Theory of Representation Learning: How Hidden Relationships Power Algorithms that can Learn without Labels

Thu, 01 May 2025 00:00:00 GMT

A Unified Theory of Representation Learning: How Hidden Relationships Power Algorithms that can Learn without Labels Hamilton, Mark T. How does the human mind make sense of raw information without being taught how to see or hear? This thesis presents a unifying theory that describes how algorithms can learn and discover structure in complex systems, like natural images, audio, language, and video - without human input. This class of algorithms has the possibility to extend our own understanding of the world by helping us to see previously unseen patterns in nature and science. At the core of this thesis’ unified theory is the notion that relationships between deep network representations hold the key discover the structure of the world without human input. This work will begin with a few examples of this principle in action; discovering hidden connections that span cultures and millennia in the visual arts, discovering visual objects in large image corpora, classifying every pixel of our visual world, and rediscovering the meaning of words from raw audio, all without human labels. In the latter half of this thesis, we will present two unifying mathematical theories of unsupervised learning. The first will explain why relationships between deep features can rediscover the semantic structure of the natural world by connecting model explainability, cooperative game theory, and deep feature relationships. The second mathematical theory will show that relationships between representations can be used to unify over 20 common machine learning algorithms spanning 100 years of progress in the field of machine learning. In particular, we introduce a single equation that unifies classification, regression, large language modeling, dimensionality reduction, clustering, contrastive learning, and spectral methods. This thesis uses this unified equation as the basis for a “periodic table of representation learning” that predicts the existence of new types of algorithms. We show that one of these predicted algorithms is a state-of-the-art unsupervised image classification technique. Finally, this work will summarize the key findings and share ongoing and future directions.

Score Estimation for Generative Modeling

Thu, 01 May 2025 00:00:00 GMT

Score Estimation for Generative Modeling Jayashankar, Tejas Kumar Recent advances in score-based (diffusion) generative models have achieved state-of-the-art sample quality across standard benchmarks. Building on the remarkable property of these models in estimating scores, this thesis presents three core contributions: 1) new objectives to reduce score estimation error, 2) a novel Bayesian-inspired optimization framework for solving inverse problems, and 3) a fast one-step generative modeling framework that is based on a novel amortized score estimation framework. In the first part of this thesis, we introduce two new score estimation objectives with applications to both implicit and diffusion-based generative models. To improve spectralbased non-parametric estimators, we propose a theoretically optimal parametric framework that learns the score by projecting it onto its top-L principal directions. Additionally, inspired by matrix-valued kernel methods, we present a second approach that lifts the score into the space of outer products, and minimizes the distance between the estimated and true scores in this higher-order space. In the second part, we shift focus from score estimation to leveraging diffusion models as data-driven priors for solving inverse problems. Centering our development around the problem of source separation, we introduce a novel algorithm inspired by maximum a posteriori estimation. This approach combines multiple levels of Gaussian smoothing with an α-posterior, enabling effective signal separation using only independent priors for the sources. We demonstrate the effectiveness of this method through its application to interference mitigation in digital communication signals. Finally, we outline how this framework can be naturally extended to tackle a broader class of inverse problems. In the final part, we return to the fundamental challenge of efficient sampling, which is critical for enabling practical data-driven engineering systems. We propose a novel generative modeling framework that enables training a one-step neural sampler from scratch. At the core of this method is a new objective based on multi-divergence minimization, guided by a novel approach for score estimation of mixture distributions. Our framework is simple to implement, stable during training, unifies several existing approaches, and achieves state-of-the-art performance in image generation tasks. Furthermore, we discuss how this framework can be naturally extended to multi-step neural sampling and adapted for fast posterior sampling—an essential component in simulation-based inverse problem solvers.

Superconducting Nanowire Integrated Circuits for Scalable Cryogenic Memory

Thu, 01 May 2025 00:00:00 GMT

Superconducting Nanowire Integrated Circuits for Scalable Cryogenic Memory Medeiros, Owen A. Superconducting nanowire integrated circuits (SNICs) are a promising class of cryogenic electronics that harness the zero resistance, high kinetic inductance, and nanoscale geometry of ultrathin superconducting wires to implement logic, memory, amplification, and sensing with minimal energy dissipation. Unlike Josephson-junction-based circuits, SNICs support compact, planar layouts compatible with single-layer fabrication and operation in unshielded cryogenic environments. This thesis develops superconducting nanowire memory (SNM) as a scalable implementation of SNICs. A modular cell architecture is introduced, exploiting hysteretic switching and inductive asymmetry to enable nonvolatile digital state storage with zero static power consumption. A hierarchical design framework is established, combining automated layout generation, electrothermal simulation in LTspice, and microscopic modeling using the time-dependent Ginzburg–Landau (TDGL) formalism. To enable scalable integration, this work implements a row–column SNM array layout and demonstrates fabrication across full 4-inch wafers using a planar, singlelayer process. Cryogenic measurements validate reliable operation in both single cells and multi-cell arrays, confirming the predictive accuracy of the design and modeling framework. Tradeoffs in bias current levels, pulse timing, and read/write conditions are systematically evaluated through cryogenic measurements, revealing their impact on bit error rate, operational margins, and energy efficiency across single cells and arrays. Together, these contributions establish SNICs as a viable and extensible platform for cryogenic memory, providing the tools, models, and infrastructure needed to enable broader adoption in quantum computing, neuromorphic systems, and other energy-constrained cryogenic applications.

Next Generation Operating Systems for the Datacenter

Thu, 01 May 2025 00:00:00 GMT

Next Generation Operating Systems for the Datacenter Fried, Joshua Modern datacenters face a fundamental challenge: handling demanding real-time and dataintensive workloads that require both microsecond-scale low latency and high throughput, while simultaneously achieving high resource utilization and efficient multi-tenancy. Traditional operating systems, designed for an era of slower hardware, introduce significant overheads to microsecond-scale I/O that prevent applications from exploiting the full performance of the underlying hardware. Furthermore, their millisecond-scale resource management is ill-equipped to handle the microsecond-level burstiness of modern workloads, leading to costly overprovisioning and idle resources. Recognizing the performance limitations imposed by traditional OSes, a common workaround has emerged: letting applications communicate directly with hardware, bypassing the OS entirely. While this approach offers performance gains by removing the OS from the critical path, existing kernel-bypass solutions require dedicated resources, extensive application rewrites, and provide weak isolation, making them unsuitable for general-purpose, shared environments. This thesis presents a new datacenter operating system, composed of three integrated systems: Shenango, Caladan, and Junction. Together, they preserve the high-performance, low-overhead I/O benefits of kernel bypass, while providing efficient resource management, strong isolation for multi-tenant workloads, and compatibility with unmodified software. First, Shenango enables applications to bypass traditional OS-mediated I/O without dedicating CPU cores solely to polling. Next, Caladan ensures that idle resources can be used productively by other applications by actively managing competition for microarchitectural resources, thereby preserving each application’s high I/O performance and responsiveness. Finally, Junction overcomes several common limitations of kernel-bypass solutions, bringing these benefits to all applications by preserving compatibility with existing software and reducing memory and polling overheads. Collectively, these systems provide the advantages of direct hardware access without sacrificing the flexibility or efficiency of a general-purpose operating system. This work demonstrates that high I/O performance, efficient resource utilization, and broad application compatibility can indeed coexist, paving the way for a new generation of datacenter operating systems.

Systematic Development of Healthcare AI: From Data Curation, Algorithm Optimization, Benchmark Design and Clinical Applications

Thu, 01 May 2025 00:00:00 GMT

Systematic Development of Healthcare AI: From Data Curation, Algorithm Optimization, Benchmark Design and Clinical Applications Gao, Mingye Artificial intelligence (AI) has brought transformative changes to healthcare industry in the recent years from various aspects, such as patient care, disease diagnosis and medical research. As healthcare systems worldwide face increasing pressure from aging populations and rising chronic disease rates, there is an urgent need for systematic approaches to develop reliable and safe AI solutions. This thesis advances the systematic development of healthcare AI through four interconnected components: data curation, algorithm optimization, benchmark design, and clinical applications. The primary contribution of this thesis focuses on establishing a comprehensive pipeline for healthcare large language models (LLMs), spanning from data curation to clinical deployment. At the data level, a rule-based filtering framework was developed to select high-quality subsets from the large pre-training corpora, significantly improving both continue pre-training and fine-tuning performance of LLMs. For safety alignment, an automated pipeline was developed for preference learning that includes preference dataset synthesis, rule-based and data-adaptive annotation, and reward model training. Additionally, two novel benchmarks were created to ensure reliability and safety of LLMs in healthcare tasks: one assessing demographic biases of LLMs across common diseases, while another assessing models’ ability to reject illogical requests from users in drug-related scenarios. Finally, LLMs were used to generating patient-friendly educational content for clinical trials, demonstrating their role in improving patient education and engagement in clinical trials. This systematic progression from data to deployment establishes a blueprint for developing safe and effective language models in healthcare settings. Beyond language models, machine learning techniques were applied on an additional healthcare task. In this project, a novel approach combining normalized cross-correlation and attention graph convolutional recurrent networks was developed to realize contactless, continuous and reliable radar-based vital signs monitoring in dynamic home environments. Through systematic data collection and algorithm optimization, the accurate heart rate can be obtained across varying radar-subject distances (2-2.5m) and subject orientations, demonstrating robust performance in real-world conditions through extensive validation in four test houses with six subjects. Collectively, these contributions advance healthcare AI development across 2 fronts: establishing frameworks for safe and effective deployment of language models in healthcare settings and enabling reliable and continuous health monitoring at-home without wearable devices.

SERenaDE: Hardware Acceleration of Cloud Serialization Frameworks

Thu, 01 May 2025 00:00:00 GMT

SERenaDE: Hardware Acceleration of Cloud Serialization Frameworks Zarkos, Christos V. Serialization frameworks are a fundamental operation of datacenters, as they enable language- and platform-neutral communication and storage. However, software serialization faces major performance bottlenecks, resulting in a significant fraction of cloud cycles dedicated to this process. Prior work has proposed specialized hardware accelerators to address these overheads. While these proposals achieve considerable speedups, they are expensive in terms of verification, fabrication, and deployment, and often hardcode too many details about the (de)serialization framework in hardware. We propose SERenaDE, a serialization framework designed to integrate general-purpose accelerators currently deployed in datacenters in order to accelerate and offload serialization to hardware. Specifically, we repurpose the Intel In-Memory Analytics Accelerator (IAA), an accelerator engine offering fast compression, to enable fast and transparent to the user serialization and deserialization, completely removing software serialization from the execution pipeline. We evaluate our system on latest-generation production machines, both with synthetic microbenchmarks, and open-source representative fleet-wide benchmarks. Our results show comparable performance in terms of per-request latency across all types of messages, while significantly improving throughput - especially at the tail -, maintaining thread scalability and achieving high compression ratios alongside substantial speedups for larger messages. Under 95th latency percentile latency constraints SERenaDE improves serialization and deserialization throughput by 13% and 30% respectively, while achieving from 0.2x to 6.94x smaller serialized message sizes for messages of a total memory layout larger than 4KB.

Modeling Biomolecular Interactions with Generative Models

Thu, 01 May 2025 00:00:00 GMT

Modeling Biomolecular Interactions with Generative Models Corso, Gabriele In 2021, DeepMind’s AlphaFold2 revolutionized single-chain protein structure prediction achieving atomic accuracy, solving a longstanding challenge in biology. However, understanding biomolecular interactions, a critical problem for advancing drug discovery and biological research, remained unsolved. This thesis presents our research to redefine the machine learning approach to this problem, modeling structures with a new generative paradigm and tailoring the neural architectures and learning tasks to the specific challenges that arose. These ideas combined with significant engineering efforts led us to develop a class of open-source models from DiffDock to the recent Boltz-1. These have significantly pushed our ability to understand biomolecular interactions, they have been widely adopted in industry and academia to help with drug development and protein design and they have opened the door to new research paradigms to push biological research further.

Designing Hardware Accelerators for Solving Sparse Linear Systems

Thu, 01 May 2025 00:00:00 GMT

Designing Hardware Accelerators for Solving Sparse Linear Systems Feldmann, Axel Solving sparse linear systems is a key primitive that sits at the heart of many important numeric algorithms. Because of this primitive’s importance, algorithm designers have spent many decades optimizing linear solvers for high performance hardware. However, despite their efforts, existing hardware has let them down. State-of-the-art linear solvers often utilize < 1% of available compute throughput on existing architectures such as CPUs and GPUs. There are many different algorithms used to solve sparse linear systems. These algorithms are diverse and often have very different computational bottlenecks. These include low arithmetic intensity, fine-grained parallellism, tight dependences, and sparsity-induced load imbalance. This thesis studies the problem of designing hardware accelerators for sparse linear solvers. We propose three novel architectures that explore different parts of the design space. The accelerators exploit static sparsity as the basis of novel hardware-software co-designed scheduling approaches. First, we introduce Spatula, an architecture designed to accelerate direct solvers. Then, we propose Azul, a hardware accelerator targeted at iterative solvers. Taken together, Spatula and Azul demonstrate significant speedups on both of the main classes of sparse linear solver algorithms. Finally, to show that our techniques are useful for end-to-end applications, we present Ōmeteōtl, an accelerator targeted at applications that use iterative solvers in their inner loop. Ōmeteōtl also shows that the techniques in this thesis generalize to sparse matrix computations beyond linear solvers. These accelerators deliver order-of-magnitude speedups over state-of-the-art GPU baselines, achieving > 100× speedups on many inputs.

Physics-Optimized Design of 3D Shapes with Part-Based Control

Thu, 01 May 2025 00:00:00 GMT

Physics-Optimized Design of 3D Shapes with Part-Based Control Zhan, Sean We introduce PhysiOPart, a computational approach for rapid generative design of 3D objects optimized for physical integrity. PhysiOPart enables users to edit and combine object parts to explore a vast design space. To model continuous surfaces of arbitrary resolution without topology restrictions, we parametrize parts with neural implicit representations. However, when parts are assembled to form an object, the resulting geometry is not guaranteed to be functional. Existing generative modeling approaches use task-specific neural predictors to approximate physical behaviors with limited accuracy. We propose an end-to-end differentiable physics simulation pipeline that performs linear static analysis to optimize for user-specified objectives, leveraging learned geometry priors. Our part-based formulation with finite element method is highly customizable, allowing for user-defined per-part materials, loads, and boundary conditions. The optimized designs exhibit improved physical behavior and can be fabricated.

Fast Assembly of Curved Structures from Flat Configuration

Thu, 01 May 2025 00:00:00 GMT

Fast Assembly of Curved Structures from Flat Configuration Zaman, Akib Imagine deploying an emergency shelter that transitions seamlessly from a flat configuration to a lifted structure, or a folded robot that is sent through a tunnel and subsequently activated to expand into a larger form at the endpoint, with a single, collective pull of strings. This scenario raises two critical questions: (i) how to decompose the structure into a flat state that encodes the 3D geometry, and (ii) where to place strings through the unit modules to achieve complete actuation. Although these questions have been explored individually, comprehensive solutions remain scarce. To address this challenge, this thesis presents a computational approach for designing freeform structures that can be rapidly assembled from initially flat configurations by a single string pull. Target structures are decomposed into rigid, spatially varied quad tiles optimized to approximate a user-provided surface, forming a flat mechanical linkage. A two-step algorithm is then applied to determine a physically realizable string path that controls only a subset of tiles, enabling smooth actuation from flat to assembled configuration. First, the minimal subset of tiles required for string control is computed by considering both the structure’s geometry and inter-tile interactions. Second, a valid string path is identified through these tiles that minimizes friction, thereby transforming the flat linkage into the target 3D form upon tightening a single string. The resulting designs can be manufactured in flat form using computational fabrication techniques: such as 3D printing, CNC milling, or molding, thereby simplifying both production and transportation. Validation is provided through a series of physical prototypes and application case studies, ranging from medical devices and space shelters to large-scale architectural installations.

Theoretical Foundations for Learning in Games and Dynamic Environments

Thu, 01 May 2025 00:00:00 GMT

Theoretical Foundations for Learning in Games and Dynamic Environments Golowich, Noah Decision-making problems lie at the heart of numerous aspects of human and algorithmic behavior across our society, ranging from healthcare systems to financial systems to interactions with the physical world. A central challenge that arises across many decision-making problems is the presence of multiple agents, often with competing incentives. To understand how agents will act in such situations, it is often productive to compute equilibria, which have the property that no agent can deviate from them and improve their utility. An additional challenge is that decisions made by agents often change the state of the environment, which is modeled as dynamic. Thus, we need efficient algorithms for learning good policies, which tell the agent what to do as a function of the environment’s state. Extensive work spanning multiple domains such as economics, computer science, and statistics has been developed to model these decision-making problems. This has led to many celebrated results, which include, for instance, a considerable body of work studying the computational properties of Nash equilibria in normal-form games, and a long line of papers on reinforcement learning. However, many of these classical works suffer from a few shortcomings: first, they often do not account for the enormous state or action spaces available to agents in realistic decision-making settings, and second, many of them do not derive computationally efficient algorithms for the desired solution concepts. These shortcomings are brought to the forefront by the remarkable recent progress in artificial intelligence, which holds promise for solving decision-making problems with enormous state or action spaces but which is often bottlenecked by computation. The objective of this thesis is to develop theoretical foundations for the computational aspects of such decision-making problems: e.g., How do we efficiently compute equilibria in large games?, and: How can we efficiently learn near-optimal policies in complex environments? Some highlights of our results are listed below—first, we study problems in which there are multiple agents and the goal is to compute some notion of equilibrium: • We show the first near-optimal rate of convergence to equilibrium for a no-regret learning algorithm in normal-form games, resolving a decade-long line of work which had aimed to establish increasingly better rates. • We establish the first algorithm with sublinear swap regret against arbitrary adversaries enjoying only polylogarithmic dependence on the number of actions, resolving a question of Blum and Mansour from 2007. • As a corollary of the preceding result, we obtain the first polynomial-time algorithm for approximating a correlated equilibrium in extensive-form games (to constant approximation error), addressing a question of von Stengel & Forges from 2008. Additionally we obtain near-optimal bounds on the communication and query complexity of approximating correlated equilibria in normal-form games (to constant approximation error), addressing several open problems in the literature. • We give the first algorithm for the sequential calibration problem with calibration error beating that of the seminal work of Foster & Vohra from 1998. Moving on to decision-making problems where the environment is modeled as dynamic (typically studied in the framework of reinforcement learning (RL)), our results include the following: • We give the first end-to-end computationally efficient algorithms for learning a nearoptimal policy in many fundamental reinforcement learning problems, such as those of (constant-action) Linear Bellman Complete MDPs and sparse linear MDPs. • We give the first quasi-polynomial time algorithm for finding a near-optimal policy in a general and well-motivated class of partially observable RL environments, and show that our bound is tight. • We prove some (perhaps surprising) hardness results that arise in multi-agent RL problems. For instance, we show that it is computationally hard to implement noregret learning algorithms in multi-agent RL environments even when the agents can coordinate on their choice of algorithm, which creates a stark contrast with simpler multi-agent learning settings (e.g., in normal-form games) where no-regret learning has formed the bedrock for a wide array of developments over the last several decades. • Nevertheless, we show that by adjusting the type of equilibrium appropriately, we can circumvent the above hardness results and derive computationally efficient decentralized algorithms for computing equilibria in multi-agent RL environments. Many of the above results have inspired follow-up work which includes applications of our results to various problems in game theory, reinforcement learning, online learning, and related domains, as well as the formulation of new problems which are inspired by the above results.

Characterizing human vision through large-scale brain imaging and computational models

Thu, 01 May 2025 00:00:00 GMT

Characterizing human vision through large-scale brain imaging and computational models Lahner, Benjamin Efforts to understand the neural underpinnings of human visual processing require sufficient experimental data and robust models. This thesis significantly contributes to both these fronts while simultaneously elucidating some of the most intriguing aspects of the human visual system. In the first chapter, I use a combination of classical machine learning, artificial neural networks, and a joint MEG/fMRI neuroimaging dataset to reveal that the human visual system extensively processes highly memorable images in regions distributed throughout visual cortex late in time. In the second chapter, I present the BOLD Moments Dataset, a large-scale fMRI dataset using short video stimuli to extend computational models of visual processing into the video domain to better understand how humans process visual content unfolding over time. The last chapter introduces a fMRI dataset aggregation framework titled MOSAIC to achieve the scale and stimulus diversity needed for training modern neural networks directly on brain responses. This body of work exemplifies how large-scale experimental data and artificial neural networks can contribute towards a robust and generalizable understanding of human visual processing.

Wireless Systems for a Sustainable Future: From Battery-Free Subsea IoT to THz-Based Agriculture Monitoring

Thu, 01 May 2025 00:00:00 GMT

Wireless Systems for a Sustainable Future: From Battery-Free Subsea IoT to THz-Based Agriculture Monitoring Afzal, Sayed Saad This thesis describes how wireless sensing can drive significant advancements in climate and sustainability. Specifically, it shows how we can leverage diverse signals—acoustics, ultrasound, THz, and optics— in unconventional ways to unlock new capabilities in underwater climate monitoring, food safety, and disaster response. The thesis introduces two novel technologies. The first technology enables long-term, ultra-low power ocean sensor networks for use in climate modeling, marine monitoring, and sustainable aquaculture. Unlike existing IoT technologies – like Bluetooth, WiFi, and GPS – which cannot work underwater, we design and implement an ultra-low power subsea backscatter communication system, enabling battery-free underwater imaging, sensing and localization. Second, the thesis describes a new technology that can support sustainability in agriculture through real-time food quality assessment that reduces food waste. In contrast to existing food quality technologies that require direct contact with produce, we introduce a new wireless system for accurate, non-invasive sensing using sub-THz signals. We describe the design, implementation, and evaluation of multiple systems that leverage these technologies to monitor the ocean and food waste: First, we present a ultra-wideband metamaterial sensor design that facilitates scalable, and long-range battery-free underwater communication. Next, we describe a system that can push the throughput of this technology using higher order modulation. Beyond building sensor networks, we demonstrate their real-world potential through two systems: one for underwater localization that uses rich spatio-temporal-spectral features for accurate positioning, and another for battery-free imaging that fuses acoustic and optical signals to capture color images in the dark. Finally, we present a novel solution for accurate fruit ripeness sensing using sub-terahertz wireless signals. These systems unlock new IoT applications in climate modeling, aquaculture, robotics, and agriculture.

Automatic Integration and Differentiation of Probabilistic Programs

Thu, 01 May 2025 00:00:00 GMT

Automatic Integration and Differentiation of Probabilistic Programs Lew, Alex K. This thesis addresses the challenge of automating fundamental operations from probability theory and calculus on probability distributions defined by higher-order probabilistic programs. It does this by developing a suite of composable program transformations for an expressive core calculus for probabilistic programming: • Integration: Compiling a probabilistic program into a deterministic representation of its expectation operator, handling potentially intractable integrals symbolically. • Unbiased estimation: Transforming programs involving intractable operations (like integration) into runnable probabilistic programs that yield provably unbiased estimates of the original value, with flexible levers for users to navigate cost-variance trade-offs. • Radon-Nikodym differentiation: Compiling probabilistic programs into implementations of a novel interface for the unbiased estimation of density ratios, of the sort that arise in Monte Carlo and variational inference. • Differentiation: Extending automatic differentiation (AD) to compose with the above transformations, enabling the optimization of expected values and density ratios of probabilistic programs. These transformations operate on an expressive higher-order probabilistic programming language and are proven correct using denotational semantics and logical relations. The resulting framework enables the sound and automated implementation of a wide range of algorithms for probabilistic inference and learning. To demonstrate the practical value of these techniques, we use them to implement three systems for scalable probabilistic inference in different domains: (1) extensions to the Gen probabilistic programming system that accelerate and automate a broad range of Monte Carlo and variational inference algorithms, (2) the PClean system for automated Bayesian reasoning about relational data, and (3) the GenLM system for controllable generation from language models. We find that our techniques enable these systems to scale to a variety of complex, real-world problems, and to achieve state-of-the-art performance on a range of benchmarks.

Minimizer-space computation

Thu, 01 May 2025 00:00:00 GMT

Minimizer-space computation Ekim, Barış C. As the volume of DNA sequencing data increases, the need for algorithmic advances to efficiently handle the data arises. One such concept is minimizers, which are genomic substrings that allow for reduced representations of larger DNA sequences. In this thesis, we introduce minimizer-space computation as a new algorithmic paradigm for DNA sequence analysis. Instead of DNA nucleotides, we consider minimizers as the letters of an extended alphabet in which algorithms operate. We present several techniques on how to efficiently construct these extended alphabets, demonstrate how to develop approaches that use these alphabets and consequently use only a fraction of sequence data, and show how fundamental biological tasks, such as genome assembly and read mapping, can be significantly accelerated over state-of-the-art methods.

Performant and Resilient Service Composition for Modern Cloud Applications

Thu, 01 May 2025 00:00:00 GMT

Performant and Resilient Service Composition for Modern Cloud Applications Li, Tianyu Modern cloud applications are often distributed systems composed from vendor-provided building blocks (e.g., object storage services, container orchestration services). Consequently, distributed fault-tolerance is a central concern for application correctness. Although each building block may offer individual fault-tolerance, the end-to-end application is still susceptible to failures, because the composition logic that orchestrates them may still fail. This thesis explores resilient composition, a systematic way to assemble fault-tolerant components into resilient end-to-end distributed applications. We begin by presenting the fail-restart system model, which captures the unique fault-tolerance challenges that arise when composing services. Based on this model, we define Composable Resilient Steps (CReSt), an atomic programming abstraction that guarantees fault-tolerance across the assembled application. We then detail efficient methods for implementing CReSt using a range of database techniques, and a novel distributed protocol that allow optimistic, speculative execution ahead of slower fault-tolerance safeguards. Together, these pieces allow developers to assemble fault-tolerant distributed systems that are correct by construction and often more performant than existing solutions.

Succinct Cryptography via Propositional Proofs

Thu, 01 May 2025 00:00:00 GMT

Succinct Cryptography via Propositional Proofs Mathialagan, Surya The goal in modern cryptography is to obtain security while minimizing the use of computational resources. In recent years, we have been incredibly successful in our pursuit for efficiency, even for cryptographic tasks that were thought to be “science fiction”. For example, we have constructions of fully homomorphic encryption and private information retrieval from standard, cryptographic assumptions which achieve the ideal levels of succinctness. However, there are still some tasks in cryptography where achieving the “ideal” efficiency from standard assumptions has evaded us. In this thesis, we study the problem of achieving succinctness in two such settings: • Can we construct succinct indistinguishability obfuscation (IO) for Turing machines? In particular, can we construct an obfuscated program whose size is independent of the input length? • Can we construct succinct non-interactive arguments (SNARGs) for all of NP? While the problems seem unrelated at first glance, the root difficulty seems to stem from a similar place: both primitives have non-falsifiable security definitions. In fact, this type of barrier exists for many other cryptographic primitives, including witness encryption. This leads to a central question which we refer to as the “non-falsifiability barrier”: how can we construct non-falsifiable primitives from falsifiable assumptions? In this thesis, we show how to leverage propositional proofs to overcome the non-falsifiability barrier, and make substantial progress in the goal of achieving succinctness in both settings. Our main result is universal construction of both SNARGs and succinct IO for Turing machines from standard assumptions using propositional proofs. We then show several applications, including rate-1 IO for many programs, the first succinct secret sharin schemes for monotone circuits, and many more. Our results establish propositional proofs as a foundational tool for achieving succinctness across a broad range of cryptographic settings.

Statistical and Algorithmic Thresholds in Spin Glasses

Thu, 01 May 2025 00:00:00 GMT

Statistical and Algorithmic Thresholds in Spin Glasses Huang, Brice This thesis studies spin glasses, disordered complex systems originating in statistical physics. Such systems model optimization, sampling, and inference problems from probability and statistics, which are of fundamental importance to modern data science. In particular, spin glasses provide natural examples of random, high-dimensional, and often highly non-convex cost or log-likelihood functions, making them an excellent testing ground for such questions. Part I of this thesis studies statistical properties of these models. Chapter 2 identifies the storage capacity of the Ising perceptron, a simple model of a neural network, subject to a numerical condition. This gives a conditional proof of a 1989 conjecture of Krauth and M´ezard. Chapter 3 gives a new proof of the celebrated Parisi formula for the free energy of the spherical mean-field spin glass, which was first proved by Talagrand and in more generality by Panchenko. Our proof takes a simpler modular approach, drawing on recent advances in spin glass free energy landscapes due to Subag. Chapter 4 characterizes the topology trivialization phase transition of multi-species spherical spin glasses and shows that lowtemperature Langevin dynamics finds the ground state in the topologically trivial regime; the latter result is new even in the single-species setting. Part II of this thesis concerns algorithms for optimization and sampling problems on spin glasses. Chapter 5 studies the problem of optimizing the Hamiltonian of a multi-species spherical spin glass. Our main result exactly characterizes the maximum value attainable by a class of algorithms that are suitably Lipschitz in the disorder. This class includes gradient-based algorithms and Langevin dynamics on constant time scales, and in particular includes the best algorithm known for this problem. This chapter is part of a series of works where we establish exact algorithmic thresholds using the branching overlap gap property (OGP), a landscape property introduced in our earlier work (which appears in our S.M. thesis). In this chapter, we develop a more robust way to establish the branching OGP that does not require Guerra’s interpolation; this allows our method to be applied to models well beyond the (single-species) mean-field spin glass we previously considered. Chapters 6 and 7 study sampling from the Gibbs measure of a spherical mean-field spin glass. Chapter 6 develops a sampling algorithm based on simulating Eldan’s stochastic localization scheme, while Chapter 7 analyzes simulated annealing of Langevin dynamics. We prove both algorithms succeed for inverse temperatures up to a stochastic localization threshold. Chapter 6 gives the first stochastic localization-based sampler with a guarantee of vanishing total variation error, improving on earlier algorithms with vanishing Wasserstein error. Chapter 7 provides the first provable guarantees for a Markov chain in this model beyond the uniqueness threshold, where mixing from worst-case initialization is provably slow.

A Cavity-Coupled Rydberg Atom Array for Quantum Science and Quantum Computing

Thu, 01 May 2025 00:00:00 GMT

A Cavity-Coupled Rydberg Atom Array for Quantum Science and Quantum Computing Hu, Beili Neutral atom arrays have rapidly emerged as a leading platform for quantum computing, boasting scalable, configurable arrays of single atoms trapped in optical tweezers, fast, high-fidelity entangling gates through Rydberg interactions, and programmable, parallelized control of qubit operations. Coupling an atom array to an optical cavity opens a new frontier. Leveraging enhanced light-atom interactions in cavity quantum electrodynamics, cavity- coupled atom arrays acquire capabilities that can further expand the neutral atom toolbox, including cavity-enhanced atom readouts, atom-photon entanglement, and photon-mediated interactions between distant atoms. This thesis presents a quantum hardware platform that integrates an array of neutral atoms with a high-finesse optical cavity. After describing the design and development of the experimental apparatus, I demonstrate high-fidelity atom state readout through the cavity, achieving improved speed and atom survival compared to conventional free-space imaging methods. I then introduce a new technique for selectively controlling atom-cavity coupling on arbitrary subsets of the array, using local AC Stark shifts on the excited states of the atoms. Building on these tools, I demonstrate fast, non-destructive cavity-based readout of atom arrays, a crucial bottleneck of atom array platforms. I also showcase real-time measurement and feedback capabilities with a demonstration of classical error correction, using a register of atomic bits. Finally, I describe progress toward implementing single- and two-qubit gates within the cavity-coupled system. By combining coherent control, tunable interactions, and high-fidelity, non-destructive readout integrated and real-time feedback, the cavity-coupled Rydberg atom array offers a promising path toward fault-tolerant quantum computing.

Estimation, Prediction and Counterfactual Inference with Dependent Observations

Thu, 01 May 2025 00:00:00 GMT

Estimation, Prediction and Counterfactual Inference with Dependent Observations Kandiros, Anthimos Vardis The success of modern data science is largely driven by access to large-scale, high dimensional data. Much of classical machine learning has been developed under the assumption that this data is generated independently from some distribution. However, this assumption is often violated when data exhibit complex dependencies across a spatial or temporal domain, or due to social interactions. In this thesis, our goal is to design and analyze methods that address these dependencies for performing three fundamental estimation tasks: unsupervised learning, supervised learning and counterfactual inference. In supervised learning, we observe a sequence of unlabeled examples and our goal is to infer some structural property from the distribution they came from. The presence of dependencies could severely complicate this question. Our results in this direction encompass both fully observable as well as latent variable models. For fully observable models, we use the celebrated Ising model to describe the dependencies. Assuming we have access to a single sample from some Ising model, which captures a variety of real-world scenarios, we design and analyze polynomial time algorithms for recovering the matrix corresponding to the network structure of the model. We then leverage these techniques to obtain improved guarantees for estimating Ising models in Total Variation (TV) distance from multiple samples. For latent variable models, we focus on the case where the structure is a tree and we get samples from the leaves, which is a common scenario in phylogenetics. Assuming the model is Gaussian, we analyze the behavior of the Expectation-Maximization (EM) algorithm, a popular heuristic for latent variable models. We show that for trees with a single latent node, EM converges to the true model and for general tree topologies, the only stationary point in the interior of the domain is the true model. We then shift our focus to discrete models and study latent tree Ising models, for which we provide polynomial time algorithms for learning the distribution of leaves in TV distance. In supervised learning, we observe a sequence of feature-label pairs and our task is to learn the predictive relationship between the features and the labels. Here, this relationship could be confounded by the presence of dependencies among labels. We formulate this question as a regression problem, where the labels of the units follow the joint distribution of an Ising model with an unknown strength parameter and external fields that are determined by the regression function. We characterize the minimax optimal rate of estimation for the various parameters and provide an efficient algorithm that achieves it. Interestingly, it might not be possible to estimate all the parameters in some cases. In counterfactual inference, we focus on the design of network experiments, where the treatment of a unit could affect the outcome of a neighboring unit in an underlying graph. Our goal is to estimate a general causal effect that can be defined as the average difference in outcomes for a unit under two different interventions. For an arbitrary such effect, we propose an experimental design, called the conflict graph design. For an unbiased estimator of that effect, we prove bounds on its variance that yield the best known rates of estimation for various effects studied in the literature, such as the average direct effect and the total effect, but also provide estimation rates for effects that have received less attention from the perspective of experimental design.

Hardening Trusted Execution Environments Against Microarchitectural Side-Channel Attacks: A Constructive Approach

Thu, 01 May 2025 00:00:00 GMT

Hardening Trusted Execution Environments Against Microarchitectural Side-Channel Attacks: A Constructive Approach Dréan, Jules Guillaume Jacques Bénony D Trusted Execution Environments (TEEs) [1–5] promised to enable secure computation even in the presence of privileged adversaries by providing hardware-enforced isolation. However, the discovery of microarchitectural side-channel and transient execution attacks [6–10] has severely undermined these security guarantees. These attacks exploit shared hardware resources and speculative execution to leak sensitive information across security boundaries, effectively bypassing the architectural isolation enforced by TEEs. The widespread impact of these vulnerabilities is evidenced by more than 43 published attacks [11] targeting commercial TEE platforms including Intel SGX, AMD-SEV, and ARM TrustZone. Existing approaches to defend against these attacks face significant limitations. Hardware-based solutions [12–14] often require complex processor modifications with significant hardware overhead. Replacing trusted hardware with cryptographic approaches incurs prohibitive performance overheads [15]. Meanwhile, formal verification methods struggle to scale to realistic code base sizes and often fail to capture subtle microarchitectural behaviors [16–18]. This thesis proposes a constructive approach to TEE security and demonstrates that practical defenses against microarchitectural attacks are achievable through careful system design. Rather than relying only on models and simulations, we focus on constructing systems that are secure by design. Our work is concretely realized through the design, implementation, and evaluation of two novel platforms: First, we present Citadel, a TEE platform that enables secure shared memory while providing precise guarantees against microarchitectural side-channel attacks. Citadel introduces relaxed microarchitectural isolation (RMI), a novel security property that allows programs to share memory while restricting information leakage to that of a non-speculative execution. To achieve RMI, Citadel combines hardware-enforced microarchitectural isolation with two simple mechanisms for controlled speculation: SpecSafe, which prevents speculative shared-memory accesses entirely, and Burst mode, which enables better performance through constrained speculation on small code snippets. Through a fully functional FPGA prototype, we demonstrate that Citadel can run real-world applications including cryptographic libraries and private ML inference with less than 5% overhead while maintaining strong security guarantees. Second, we develop Argos, an “integrity-only” TEE specifically designed for verifiable fully homomorphic encryption, that enables the deployment of FHE schemes in real-world settings where malicious security is required. We show that by carefully constraining the attack surface and employing simple hardware mechanisms, we can achieve complete security against microarchitectural attacks. Argos introduces a simplified transcript-based attestation scheme that only requires one signature per FHE computation, amortizing the cost of relying on a physical TPM to microarchitecturally isolate secrets. Argos can be used to not only enforce circuit-level integrity of FHE schemes but can also be extended to support more complex FHE-based applications that take (potentially poisoned) input from the (malicious) circuit evaluator. Argos is compatible with commodity hardware and only incurs minimal performance overhead with an average of 3% overhead for FHE evaluation and 8% overhead for complex protocols. Through these systems, we show that effective defenses can be built against microarchitectural side channel and transient execution attacks. Our constructive approach yields practical systems that are secure by design while maintaining efficiency and usability. This thesis opens new possibilities for the deployment of trusted hardware by demonstrating concrete paths toward robust microarchitectural security.

Steering Robots with Inference-Time Interactions

Thu, 01 May 2025 00:00:00 GMT

Steering Robots with Inference-Time Interactions Wang, Yanwei Imitation learning has driven the development of generalist policies capable of autonomously solving multiple tasks. However, when a pretrained policy makes errors during deployment, there are limited mechanisms for users to correct its behavior. While collecting additional data for finetuning can address such issues, doing so for each downstream use case is inefficient at deployment. My research proposes an alternative: keeping pretrained policies frozen as a fixed skill repertoire while allowing user interactions to guide behavior generation toward user preferences at inference time. By making pretrained policies steerable, users can help correct policy errors when the model struggles to generalize—without needing to finetune the policy. Specifically, I propose (1) inference-time steering, which leverages user interactions to switch between discrete skills, and (2) task and motion imitation, which enables user interactions to edit continuous motions while satisfying task constraints defined by discrete symbolic plans. These frameworks correct misaligned policy predictions without requiring additional training, maximizing the utility of pretrained models while achieving inference-time user objectives.

Advancing Deep Learning Efficiency: From Specialized Co-Design to Automated Generation

Thu, 01 May 2025 00:00:00 GMT

Advancing Deep Learning Efficiency: From Specialized Co-Design to Automated Generation Lin, Yujun The explosive growth of artificial intelligence (AI) technologies, particularly large-scale deep learning models such as large language models and diffusion models, has intensified the demand for efficient full-stack inference solutions that effectively balance performance and costs. This work will present a comprehensive exploration into the algorithm-system co-optimization, hardware design specialization and automation for scalable AI deployment. First, we begin with algorithmic optimization for large-scale models, including large language models and diffusion models, developing inference libraries that leverage quantization to boost the performance of generative AIs on existing GPU platforms. Next, we design specialized hardware accelerators for domain-specific applications, specifically point cloud understanding, emphasizing efficiency improvements through the exploitation of data sparsity. Finally, we open up the hardware design space beyond template-based sizing, and progress into the automated learning-based co-design of neural network and hardware architectures, maximizing their synergy with a full-stack joint optimization. We then introduce an automated framework for spatial accelerator generation, transforming high-level mappings into custom hardware designs that support scalable deployment. Together, these contributions advance AI inference efficiency by bridging the gap between advanced computational requirements and hardware capabilities, between theoretical potential and practical solutions, and between design cost and effectiveness.

Learning Theoretic Foundations for Understanding Quantum Systems

Thu, 01 May 2025 00:00:00 GMT

Learning Theoretic Foundations for Understanding Quantum Systems Liu, Allen Understanding and harnessing the power of quantum systems has the potential to transform many domains in science and technology. However, before we can achieve these aspirations, we must first build a better understanding of how quantum systems fundamentally behave. In this thesis, we approach this question through the lens of learning theory to develop new paradigms for learning about quantum systems and understanding their structural properties. We deliver several surprising results, upending previous beliefs about even fundamental laws and giving provably efficient algorithms for learning about quantum systems in settings previously conjectured to be intractable. Typically in quantum many-body systems, the particles in the system interact locally with respect to some geometry as described by a local Hamiltonian. Two key questions are first, understanding equilibrium properties of a system with a given Hamiltonian and second, recovering the Hamiltonian from measurements of the properties of the system. For the first, we prove a universal law that there is a sudden death of entanglement, at a critical temperature depending only on the geometry but not on the system size. For the second, we give the first efficient algorithm for recovering the Hamiltonian at any temperature, breaking a conjectured barrier at low temperatures. Beyond systems with local interactions, we also consider learning and testing properties of general quantum states, focusing on the interplay between statistical complexity and near-term quantum device constraints, only allowing for entangled measurements over a limited number of copies of the state. We characterize the optimal rates for learning and testing with single-copy measurements and for multi-copy measurements in many relevant near-term regimes.

Domain Wall Based Magnonics in Iron Garnet

Thu, 01 May 2025 00:00:00 GMT

Domain Wall Based Magnonics in Iron Garnet Gross, Miela J. Magnonic devices leverage magnons, quantized spin waves, as the mechanism to process and transfer information. In materials with low Gilbert damping, these spin wave-based systems enable ultra-fast operation while eliminating thermal heating and leakage currents inherent to conventional electron-based microelectronics. To maximize energy efficiency and processing speed, materials like iron garnets, ferrimagnetic insulators with tunable magnetic properties, are essential. Key magnetic parameters, including saturation magnetization, perpendicular magnetic anisotropy, coercivity, and Gilbert damping, can be tailored through elemental substitution or strain engineering in thin films. Furthermore, relativistic domain wall velocities reported in yttrium iron garnet (YIG), bismuth substituted YIG, and thulium iron garnet lay the foundations for high-speed operation. These unique attributes position garnets as ideal materials for the development of magnonic devices that integrate efficiency, speed, and versatility. This thesis presents my research on integrating thin film garnets into a domain wall based magnonic devices. It begins by exploring the magnetic characterization of thin film iron garnets, including the growth process, temperature dependent magnetic behavior, and tunable magnetic anisotropy. Next, we report on magnonics within the garnet, focusing on the interactions between spin waves and domain walls. Finally, we demonstrate a write mechanism for a magnonic device driven by spin wave-induced domain wall motion, providing detailed characterization of the device behavior and performance. These results underscore the potential of iron garnets for magnonic-based device applications and offer insights into the efficiency of write mechanism, paving the way for energy-efficient high-speed spintronic technologies.

Mitigating Inhomogeneity in High-Field MRI Excitations: Arbitrary Waveform Optimization and Multiphoton Parallel Transmission (MP-pTx)

Thu, 01 May 2025 00:00:00 GMT

Mitigating Inhomogeneity in High-Field MRI Excitations: Arbitrary Waveform Optimization and Multiphoton Parallel Transmission (MP-pTx) Drago, John M. High-field magnetic resonance imaging (MRI) using a standard volume coil results in a spatially varying flip angle across the body, which renders images difficult to clinically interpret. This arises from the complex interactions of electromagnetic fields from current-carrying elements surrounding the imaging region. Parallel transmission (pTx) mitigates this issue by employing multiple high-power, independently controlled transmit elements for more precise excitation control. However, since the wavelength of the applied radio waves is shortened in tissue, the effect becomes highly dependent on the patient’s anatomy. As a result, optimization must be performed on a patient-by-patient basis, and methods that attempt full control of these independent waveforms are too computationally intensive to execute during the limited examination time. Additionally, the high-field excitations create complex electric field distributions that require control and careful monitoring to avoid excessive tissue power deposition (and ultimately heating), quantified as the specific absorption rate (SAR). To address these challenges, we introduce a method for optimizing patient-specific pulses using a global waveform (Ritz) approach, enabling rapid, in-scanner optimization. While pTx effectively addresses flip angle inhomogeneity, it remains costly and introduces challenges in SAR management. We address the SAR management and cost problems of pTx by introducing and characterizing the MP-pTx method, which leverages the multiphoton phenomenon to improve homogeneity using a standard volume coil supplemented with low-frequency (kilohertz) parallel channels. MP-pTx reduces costs and simplifies SAR management by shifting the parallel irradiation to low-cost, lowSAR shim array channels. These channels supplement an off-resonant excitation from a conventional birdcage coil with an oscillating, z-directed field that satisfies the resonance condition for spin state transitions.

Generative Latent Motion Planning and Reinforcement Learning for Legged Locomotion

Thu, 01 May 2025 00:00:00 GMT

Generative Latent Motion Planning and Reinforcement Learning for Legged Locomotion Miller, Adam Joseph In recent years, reinforcement learning has demonstrated its promise as a powerful tool for developing innovative and advanced control systems for legged robots. The method’s robustness, versatility, and generality have made it a prime candidate for future robotic systems deployed in the real world. Through the development of more advanced machine learning algorithms and more reliable and efficient physics simulators, reinforcement learning continues to improve and enable new, dynamic, and agile capabilities. While the results are often impressive and the tools relatively beginner-friendly, there remain impediments to scalable and reliable progress. Poor reward function scaling, challenges balancing exploration versus exploitation, and misalignment from the engineer’s intent are roadblocks to better performance. To get beyond these limitations, new tools and frameworks are necessary. In this work, I present novel methods to address these challenges and extend the capabilities of reinforcement learning on robot hardware. Through the quantification of the distributional sim-to-real gap, simulation model optimization for hardware matching, latent space motion sequence planning, and latent style training, I demonstrate never-before-seen performance on legged hardware.

Learning from Weak Supervision: Theory, Methods, and Applications

Thu, 01 May 2025 00:00:00 GMT

Learning from Weak Supervision: Theory, Methods, and Applications Lang, Hunter The growing demand for high-quality labeled data to train machine learning models has driven widespread adoption of weak supervision and synthetic data methods, which use automated models instead of humans for annotation. Large language models (LLMs) have further accelerated this trend because their zero- and few-shot classification performance enables them to serve as effective “synthetic annotators” for various tasks. In practice, the data generated by these weak annotators is imperfect, but it enables the training of strong models. However, theoretical understanding of why training one model on the outputs of another leads to strong performance remains limited, especially when the annotator model exhibits suboptimal performance on the target task. In this thesis, I develop a theoretical framework for learning from weak supervision that captures the key aspects of the problem better than existing approaches in the crowdsourcing and learning-with-noisy-label literature. This framework establishes structural conditions that explain when and why weak supervision can reliably train strong models. Building on these theoretical results, the second part of the thesis introduces methods to improve how models learn from weak supervision and applies these methods to low-labeled-data settings.

Multi-fidelity Optimal Trajectory Generation: Optimal Experiment Design for Robot Learning

Sun, 01 Sep 2024 00:00:00 GMT

Multi-fidelity Optimal Trajectory Generation: Optimal Experiment Design for Robot Learning Ryou, Gilhyun Data-driven methods have significantly advanced robot learning, yet their direct application to real-world robots remains challenging, particularly under extreme conditions. This challenge is especially pronounced for highly maneuverable vehicles like quadrotor aircraft, which often operate in scenarios requiring rapid maneuvering, such as racing, defense systems, or safety-critical obstacle avoidance. In such extreme conditions, real-world constraints like control delays, state estimation errors, and battery voltage fluctuations often compromise trajectory reliability, even when conforming to ideal dynamics. However, the typical data-driven methods are usually developed in simulated environments. Consequently, the transition to real-world dynamics requires extensive fine-tuning, which can be risky, as perfect training in simulations does not guarantee safe transitions to real-world dynamics. This thesis employs methods from optimal experiment design to address these challenges. By quantifying uncertainty and maximizing information gain, the approach aims to safely and efficiently design the real-world experiments required for accurate constraint modeling. In the first chapter, we present a multi-fidelity Bayesian optimization method that searches for time-optimal speed profiles for quadrotor aircraft, effectively balancing numerical simulations with real-world flight experiments. The second chapter extends the optimal experiment design method to a high-dimensional online planning problem through integration with reinforcement learning. The proposed algorithms, trained and validated through real-world flight experiments, significantly outperform baseline methods in trajectory time and computational efficiency. Additionally, these algorithms have been adapted to various planning problems, including fixed-wing aircraft planning, cooperative multi-drone systems, and energy-efficient trajectory generation.

Efficient Systems for Large-Scale Graph Representation Learning

Thu, 01 May 2025 00:00:00 GMT

Efficient Systems for Large-Scale Graph Representation Learning Huang, Tianhao Graph representation learning has gained significant traction in critical domains including finance, social networks, and transportation systems due to its successful application to graphstructured data. Graph neural networks (GNNs), which integrate the power of deep learning with graph structures, have emerged as the leading methods in this field, delivering superior performance across diverse graph related tasks. However, training graph neural networks on large-scale datasets encounters scalability challenges on current system architectures. First, the sparse, non-localized structures of real-world graphs lead to inefficiencies in data sampling and movement. This characteristic heavily stresses system input/output (I/O), particularly burdening the peripheral buses during the sampling phase of GNN training. Second, the suboptimal mapping of training procedure to GPU kernels leads to compute inefficiencies, including substantial kernel orchestration overhead and redundant computations. Addressing these challenges requires a comprehensive, full-stack optimization approach that fully leverages hardware capabilities. This thesis presents two complementary works to achieve the goal. The first work, Hanoi, unblocks the data loading bottleneck in out-of-core GNN training by co-designing the sampling algorithms to align with the hierarchical memory organization of commodity hardware. Hanoi drastically reduces I/O traffic to external storage, delivering up to 4.2× speedup over strong baselines with negligible impacts on the model quality. Notably, Hanoi is able to obtain competitive performance close to in-memory training with only a fraction of memory requirements. Building on this foundation, the second work, Joestar, introduces a unified framework for optimized GNN training on GPUs. Joestar adapts the multistage sampling approach from Hanoi to in-memory training which frees CPUs from heavy data loading workloads. Joestar also identifies novel kernel fusion opportunities and formulates better execution schedules by jointly considering the sampling and compute stages. Combined with compiler infrastructure in PyTorch, Joestar achieves state-of-the-art GNN training throughputs for billion-edge graph datasets on a single GPU.

Generalizable Long-Horizon Robotic Manipulation under Uncertainty and Partial Observability

Thu, 01 May 2025 00:00:00 GMT

Generalizable Long-Horizon Robotic Manipulation under Uncertainty and Partial Observability Curtis, Aidan A central goal in embodied artificial intelligence is to enable autonomous agents to accomplish complex, long-horizon tasks in novel, partially observable environments. In these scenarios, agents must effectively reason about uncertainty, generalize from limited experiences, and proactively plan actions to acquire missing information. This thesis tackles these core challenges by developing and evaluating novel methods specifically designed for partially observable contexts. The first part of this thesis introduces an enhanced heuristicguided planning technique that increases search efficiency in sparse-reward domains with significant uncertainty. Next, we investigate how symbolic reasoning can be integrated into the decision-making framework, accelerating search through the use of temporal and belief-space abstractions. Next, we propose a method for sequencing low-level reinforcement learning skills alongside information gathering actions, enabling increased task complexity and robustness in real-world tasks. Lastly, we show how large language models may be leveraged for few-shot model learning, allowing agents to rapidly adapt and generalize to new scenarios. The methods presented in this thesis advance the state-of-the-art in embodied AI by enabling robots to better handle uncertainty and incomplete information, ultimately paving the way for more capable, exploratory, and risk-aware autonomous systems.

Graph-based Vector Search Algorithms for Retrieval-Augmented AI Systems

Thu, 01 May 2025 00:00:00 GMT

Graph-based Vector Search Algorithms for Retrieval-Augmented AI Systems Zhang, Ziyu The recent advancement of large language models (LLMs) and large multimodal models (LMMs) greatly enhances the capabilities of AI systems such as recommendation systems and coding assistants, making them more practical for real-world deployment. However, these models cannot directly interact with large volumes of data in a knowledge corpus during inference/task time due to inherent architectural limits and cost concerns. Encoding data into vector embeddings and leveraging approximate nearest neighbor search (ANNS) have thus become an important data processing primitive in AI systems following the introduction of retrievel-augmented generation (RAG). However, the complexity of tasks these AI systems aim to solve introduces challenges for existing ANNS algorithms. I developed methods to expand existing ANNS algorithms to address two such challenges: freshness and heterogeneity in the data. Graph-based ANNS algorithms have been proven to have superb cost versus approximation quality trade-off yet follow a simple intuition of best-first search. I focus on adapting graph-based ANNS algorithms to two settings featuring emerging challenges. (1) Data is updated constantly. Existing algorithms are inefficient under deletions and not robust against different orderings in the workload. I propose methods addressing these problems and developed an algorithm supporting updates effectively and efficiently based on Vamana, a state-of-the-art graph-based ANNS algorithm. (2) Data is heterogeneous in format, modality, and how they relate to a query, making the similarity difficult to capture by the canonical ANNS definition. I explore ways to model the similarity between heterogeneous sources and using graph-based ANNS approaches to perform semantic search in this setting. I test this approach under an end-to-end multimodal question-answering system developed in-house.

Adaptive Abstractions for Robust Hierarchical Manipulation Planning

Thu, 01 May 2025 00:00:00 GMT

Adaptive Abstractions for Robust Hierarchical Manipulation Planning Noseworthy, Michael S. In this thesis, we address the problem of long-horizon robotic manipulation under partial observability. Tasks such as gearbox assembly or tidying a workstation involve many objects and necessitate a variety of manipulation capabilities. These long-horizon tasks are commonly addressed by hierarchical approaches, which introduce state and action abstractions to make planning tractable. However, our abstractions often rely on imperfect models of the world, which can lead to brittle execution. Furthermore, these abstractions depend on having accurate state information, which is often only noisily sensed, if sensed at all. For example, in the assembly domain, the pose of each part may only be known within a few millimeters, and a box’s mass distribution may be completely unsensed. To deploy robots outside of structured environments like the factory, they will need to be robust to model misspecification and partial observability. The central idea of this thesis is that we can develop adaptive abstractions to improve the robustness of hierarchical planning once the robot is deployed. Adaptive abstractions incorporate observations from the real world that are informative about misspecifications and partial observability, essentially allowing the planner to adapt to its deployment environment. We explore this idea by developing three types of models that enable this adaptivity at different levels of the abstraction hierarchy: plan feasibility models, adaptive samplers, and reactive control policies. In our first contribution, we consider adding adaptivity to a task and motion planning system at the task-planning level. We focus on a setup where the robot has access to a set of parameterized skills, but these skills are derived from imperfect models. To enable robust planning, we propose to autonomously learn skill feasibility models once the robot is deployed through a curious exploration phase. Critically, we propose a novel active learning framework to enable efficient learning without human intervention. We show that the resulting feasibility model leads to robust task performance on multiple downstream tasks in a stacking domain. Our second contribution looks at developing adaptive samplers that can incorporate information about object state that is typically unobserved (e.g., inertial and frictional properties). General-purpose belief representations can handle this partial observability, but online inference is computationally expensive. Instead, we propose to use an offline phase to learn an inference network that directly predicts a distribution over object properties that is consistent with the interaction history. We show that inference networks enable efficient adaptation in a grasping domain with heavy objects. Our final contribution focuses on learning adaptive controllers such that robustness is handled at the lowest level of the abstraction. We consider precise contact-rich manipulation tasks that are sensitive to pose estimation errors. To overcome noisy poses at the control level, explorative contact is necessary, but unintended forces can lead to catastrophic outcomes such as part slippage or damage. We propose to use simulation in an offline phase to train reactive force-aware policies. The policies are trained to overcome pose uncertainty while using force-sensing to adaptively limit excessive forces. The result is robust real-world performance on the multistage assembly of a planetary gearbox system, which includes insertion, gear-meshing, and nut-threading tasks. In summary, adaptive abstractions can be used to increase the robustness of hierarchical manipulation planning, an important step in deploying robots outside of the lab or factory. Throughout the thesis, we validate the proposed approaches on the real robot in stacking and assembly domains.

Methods for Generalization Under Distribution Shift

Thu, 01 May 2025 00:00:00 GMT

Methods for Generalization Under Distribution Shift Netanyahu, Aviv Machine learning systems have achieved remarkable performance in tasks where test data closely resembles the training distribution. However, real-world applications often require systems capable of handling more challenging situations -- specifically, adapting to new tasks and extrapolating to data points outside the distribution of the training set. The current paradigm for handling distribution shifts is collecting and training models on large datasets. This work offers two more principled frameworks that enable machine learning models to generalize effectively to out-of-distribution scenarios without sacrificing the power of modern overparameterized models. The first framework converts an out-of-support zero-shot generalization problem into an out-of-combination problem via a transductive reparameterization, which is possible under low-rank style conditions. We explore how this idea can be applied to domains like robotics, where the environment is changing, and materials and molecular design, where predicting properties of materials or molecules outside of known ranges is crucial to driving more efficient materials discovery. The second framework focuses on few-shot task learning, which involves agents learning new tasks from minimal data and applying them to new environments. We formulate the problem of few-shot task learning as Few-Shot Task Learning through Inverse Generative Modeling, which allows us to leverage the power of neural generative models pretrained on a set of base tasks. We adapt a method for efficient concept learning to few-shot task learning based on our formulation and rapidly learn new tasks with only a few examples, enabling task execution from autonomous driving to real-world robotic manipulation tasks in novel settings without the need for extensive retraining.

Scaling 3D Scene Perception via Probabilistic Programming

Thu, 01 May 2025 00:00:00 GMT

Scaling 3D Scene Perception via Probabilistic Programming Gothoskar, Nishad Understanding and interpreting the 3D structure of the world is a central challenge in artificial intelligence. Our physical world is 3D, yet our AI systems often “see” that world through pixels and images. In order to build truly intelligent AI systems, we must go beyond pixels and images and build 3D vision systems that can build meaningful and useful 3D representations of the world. This is the problem of 3D scene perception. How do we transform raw visual input into 3D representations of the world? 3D scene perception has numerous applications from robotics to augmented reality. Despite the advances over the last decade, 3D perception remains a major bottleneck in real-world robotics applications. The challenge stems from the immense variability in real-world conditions, e.g. lighting, color, viewpoint, camera properties, object appearance, the incompleteness of visual data due to limited resolution, noise, and occlusions, and the approximations in our models of visual data. Developing more robust and generalizable 3D perception systems would be an important step towards more general-purpose robotics. In this thesis, we explore a probabilistic architecture for 3D perception based on structured generative models and probabilistic programs. We begin with 3DP3, the first iteration of our approach, which infers 3D scene graphs from real-world depth image data. 3DP3 demonstrates that our method could work on real-world benchmarks and correct commonsense errors from deep learning systems. Building on this foundation, we develop Bayes3D, which scaled up these ideas using a GPU-accelerated image likelihood and generative model alongside a parallel coarse-to-fine inference algorithm. Next, we explore two approaches for incorporating RGB image data into generative 3D graphics programs, expanding their applicability. We then introduce DurableVS, which extends inverse-graphics techniques to model scenes involving a robot and multiple cameras, enabling precise control of a robot. Finally, we present Gen3D, which integrates all the key ideas from this thesis into a real-time 3D perception system that uses multi-resolution probabilistic models of 3D matter to enable real-time tracking that is competitive with vision transformers and 3D Gaussian splatting, state-of-the-art methods in computer vision and computer graphics.

Efficient Generative Models for Visual Synthesis

Thu, 01 May 2025 00:00:00 GMT

Efficient Generative Models for Visual Synthesis Yin, Tianwei While current visual generative models produce high-quality outputs, they suffer from significant computational costs and latency, limiting their applicability in interactive settings. In this dissertation, we introduce a suite of techniques designed to enhance the efficiency of generative models for image and video synthesis. First, we propose distribution matching distillation, a method that enables the training of one- or few-step visual generators by distilling knowledge from computationally expensive yet highly capable diffusion models. Next, we develop improved distillation techniques that enhance robustness and scalability, culminating in a production-grade few-step image generator. This system is now deployed in widely used software, generating hundreds of millions of images annually. Finally, we extend our approach to video generation by adopting an autoregressive paradigm, significantly reducing latency and enabling fast interactive video generation and world simulation.

Characterization of pGaN-gate power HEMTs

Thu, 01 May 2025 00:00:00 GMT

Characterization of pGaN-gate power HEMTs Yu, Yue This thesis presents a comprehensive study of p-GaN gate GaN High Electron Mobility Transistors (HEMTs) with a focus on understanding how fabrication process variations and gate structural designs impact key electrical performance metrics. Five industry-fabricated wafers, each processed with distinct etch depths, contact strategies, and p-GaN surface configurations, were characterized using a combination of DC and pulsed I–V measurements. Full-transistor modules were evaluated alongside specialized test structures to enable both system-level and localized analysis. DC measurements using the Keysight B1505A system revealed that more aggressive gate contact schemes improved ON-resistance and transconductance, but often at the cost of increased gate leakage and reduced threshold control. Pulsed-IV characterization with the Auriga AU4750 system uncovered dynamic Ron degradation behavior and charge trapping effects, especially under high drain bias conditions. Extracted time constants demonstrated process-dependent trends, with wafers retaining more of the p-GaN surface exhibiting slower charge detrapping and more severe transient effects. Specialized test structures provided additional insights into gate lateral conduction, sheet resistance, and contact asymmetry, reinforcing the connection between device layout, processing, and observed variability. These findings highlight critical trade-offs in the design and fabrication of p-GaN gate GaN HEMTs and offer design-aware strategies for optimizing performance and reliability.

Contactless Sleep and Physiological Monitoring Using Artificial Intelligence and Radio Waves

Thu, 01 May 2025 00:00:00 GMT

Contactless Sleep and Physiological Monitoring Using Artificial Intelligence and Radio Waves He, Hao Remote monitoring of sleep and physiological signals is critical for tracking human health, managing diseases, and enabling early intervention. However, existing monitoring solutions face two major limitations: (1) they are often unsuitable for vulnerable populations—such as infants and seniors—and (2) most of them raise concerns about measurement accuracy. We propose a novel, contactless approach that addresses both challenges by combining advances in artificial intelligence (AI) and radio-frequency (RF) sensing. Our solution makes monitoring more comfortable, accessible, and affordable, while still delivering clinically meaningful insights. This thesis makes four fundamental contributions: First, we introduce a system that can extract high-fidelity breathing signals from ambient RF reflections, even in complex scenarios where multiple individuals are present, such as couples sharing a bed. Second, we develop an AI-based sleep monitoring framework that generates sleep hypnograms and detects respiratory events entirely without the need for on-body sensors. Third we develop AI models that infer critical biomarkers—such as blood oxygen saturation (SpO₂) and inflammation (C-reactive protein levels)—in a fully passive and non-intrusive manner. Finally, inspired by the success of large language models, we show that physiological signals can be represented and interpreted analogously to language. This insight enables effective translation between modalities (e.g., from respiration to EEG) and unlocks robust representation learning for downstream clinical tasks. Together, these contributions establish a new paradigm for remote sleep and physiological monitoring—one that is contactless, continuous, and passive. We validate our system on real world datasets and demonstrate its potential to fundamentally transform clinical care and home health monitoring.

Crystallization of Glauber's salt

Wed, 01 Jan 1936 00:00:00 GMT

Crystallization of Glauber's salt Coberly, C. Wheeler. Thesis: M.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1936; Includes bibliographical references (leaf 39).

Investigation of torquemeters for high speed shafts

Thu, 01 Jan 1959 00:00:00 GMT

Investigation of torquemeters for high speed shafts Saluja, Narinder S. (Narinder Singh) Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1959; Includes bibliographical references (leaves 64-67).

The resonant-frequency shift of a microwave cavity caused by the high-density plasma in semiconductors, as a function of magnetic field

Thu, 01 Jan 1959 00:00:00 GMT

The resonant-frequency shift of a microwave cavity caused by the high-density plasma in semiconductors, as a function of magnetic field Weber, Robert. Thesis: Sc. D., Massachusetts Institute of Technology, Department of Physics, 1959; Includes bibliographical references (leaves 46-47).

Analysis of angular scintillation of radar echoes

Tue, 01 Jan 1952 00:00:00 GMT

Analysis of angular scintillation of radar echoes Graham, James William. Thesis: M.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1952

Rapid transit use of existing rail lines

Wed, 01 Jan 1958 00:00:00 GMT

Rapid transit use of existing rail lines Kenyon, Michael D. Thesis: B.S., Massachusetts Institute of Technology, Department of Civil Engineering, 1958; Includes bibliographical references (leaf 25).

Mass transfer from rotating cylinders

Sun, 01 Jan 1956 00:00:00 GMT

Mass transfer from rotating cylinders Cotter, John.; Schmidt, Guy L. Thesis: B.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1956; Bibliography: leaf 38.

An observation about the Chicago Council and its policies

Thu, 01 Jan 1948 00:00:00 GMT

An observation about the Chicago Council and its policies Naber, Fred P. Thesis: B.S., Massachusetts Institute of Technology, Department of Business and Engineering Administration, 1948

The design and construction of an ultra-high vacuum field-ion microscope.

Thu, 01 Jan 1970 00:00:00 GMT

The design and construction of an ultra-high vacuum field-ion microscope. Olson, Gregory Bruce. Thesis: M.S., Massachusetts Institute of Technology, Department of Metallurgy and Materials Science, 1970; Bibliography: leaf 35.

South End Center for the Arts.

Fri, 01 Jan 1965 00:00:00 GMT

South End Center for the Arts. Dunbar, Gary Arthur. Thesis: B. Arch., Massachusetts Institute of Technology, Department of Architecture, 1965; "Special requirements for group A occupancy: theatres" leaves [34-42] inserted. "Special requirements for group C occupancy: schools" leaves [50-54] inserted.; Bibliography: leaf 20.

Analysis of braced excavations.

Fri, 01 Jan 1971 00:00:00 GMT

Analysis of braced excavations. Wong, Ing Hieng. Thesis: Sc. D., Massachusetts Institute of Technology, Department of Civil Engineering, 1971; Three leaves on transparent sheets. Vita.; Bibliography: leaves 95-99.

Shipleasing as a prospective method of l/t financing for international shipowners.

Tue, 01 Jan 1974 00:00:00 GMT

Shipleasing as a prospective method of l/t financing for international shipowners. Angelicoussis, John Anthony. Thesis: M.S., Massachusetts Institute of Technology, Sloan School of Management, 1974; Includes bibliographical references.

Modelling rail freight management.

Sun, 01 Jan 1978 00:00:00 GMT

Modelling rail freight management. Assad, A. (Arjang) Thesis: Ph. D., Massachusetts Institute of Technology, Sloan School of Management, 1978; Vita.; Bibliography: leaves 277-292.

The effects of rapid thermal annealing on gallium arsenide grown by MOCVD on silicon substrates

Fri, 01 Jan 1988 00:00:00 GMT

The effects of rapid thermal annealing on gallium arsenide grown by MOCVD on silicon substrates Lehman, LeNore Louise. Thesis: M.S., Massachusetts Institute of Technology, Department of Materials Science and Engineering, 1988; Includes bibliographical references.

Acoustic-phonetic and lexical constraints in word recognition: lexical access using partial information

Sun, 01 Jan 1984 00:00:00 GMT

Acoustic-phonetic and lexical constraints in word recognition: lexical access using partial information Huttenlocher, Daniel P. Thesis: M.S., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 1984; Bibliography: leaves 73-77.

Metabolism in vivo of 1, 3-butanediol in the rat

Fri, 01 Jan 1971 00:00:00 GMT

Metabolism in vivo of 1, 3-butanediol in the rat Nahapetian, Aratoonnaz, author. The metabolism of 1, 3-butanediol (BD) was investigated in vitamin B 12 -deficient and normal rats and in liver slice and diaphragm systems. Body weight gain and feed efficiency were determined in rats fed ad libitum for five weeks on a basal 5% BD or 5% sodium propionate diet with and without vitamin B12. The rats were train-fed for ten months on the same diets. The presence of sodium prop i onate in vitamin B12-deficient basal diets resulted in reduced food intake while BD had the opposite effect. As a result, vitamin B12-deficient rats fed a 5% sodium propionate diet grew less than those fed a 5% BD diet. The metabolism in vivo of BD labeled in carbon-1 (BD-l-cl4) and carbon-4 (BD-4-cl4) were compared to the metabolism of propionate-l-cl4 (PRP-l-cl4) in vitamin B12-deficient and normal rats. Vitamin B12 deficiency reduced the oxidation of sodium propionate but not that of BD, and had no effect on glycogen labeling from BD-l-cl4 and BD-4-cl4. For PRP-l-cl4 however, vitamin B12 deficiency resulted in not only no incorporation of label but liver glycogen levels were very small. On the other hand , when vitamin B12 was present in the diet, the labeling of glycogen from propionate was higher than that from either of the BD-labeled test compounds. Methylmalonic aciduria and urinary loss of ingested activity was higher in vitamin B12-deficient rats fed PRP-l-cl4 than in those fed l abe l ed BD. Nearly all of the urinary activity of vitamin B 1 2-deficient rats fed PRP-l-cl4 was in the form of me t hy l malonic acid (MMA), while little, if any, of the activity was found in the MMA fraction of urine of vita m in B12-deficient rats fed labeled BD. The metabolism in vivo of BD-c14 and BD-3-c14 was investigated in normal rats. About eighty percent of BD was oxidized to carbon dioxide within 32 hours. Its oxidation in the first eight hours was higher when BD was administered intraperitoneally than when it was fed by stomach tube. The loss of ingested activity in the urine expressed as a percentage of total intake and 1,3-BD was higher at the higher doses of BD. However, the activity in urinary BD could not account for all the activity in the urine. A considerable amount of ketone bodies was detected in urine of rats after feeding BD while no detectable ketone bodies were found in the urine of control rats. In addition, relative specific activities of urinary BD and S -hydroxybutyrate were 0.91 and 0.50 respectively. Polarimetry of both purified urinary BD and S -hydroxybutyrate showed that the percentages of (+)- and (-)-isomers of both compounds were 40 and 60% respectively. The metabolism in vitro of BD-3-c14 and DL-S - hydroxybutyrate-4-cl4 were investigated in systems which contained liver slices alone, diaphragm alone or both liver slices plus diaphragm. The oxidation rate of S -hydroxybut y rate was lower in liver slices than in either the diaphragm or the liver slices plus diaphragm systems. Moreover, the rate of oxidation of S -hydrox y- butyrate was highest in the system which included both liver slices and diaphragm. On the other hand, the oxidation rate of BD was lower in the system which had only diaphragm than in the other two systems. However, the rate of BD oxidation was highest in the system - wh ich included both liver slices and diaphragm. Presence of BD gave rise to increased D-(-)- S -hydroxybutyrate and acetoacetate in systems which contained liver slices or liver slices plus diaphragm. In addition, the production rate of D-(-)- S - hydroxybutyric acid was higher than that of acetoacetate in the pre sence of BD, while the opposite was true in its absence. Finally, all the radioactivity in the control incubation media was accounted for by BD-3-cl4, while about 1.5 and 98.5 percent of incubation media activity were recovered in S -hydroxybutyrate and BD peaks, respectively, in incub at ion systems containing liver. The results of this study indicate that 1,3-BD and sodium propionate do not share a common metabolic pathway in the rat. The data suggest, however, that nahapetian-4-1, 3-BD is most probably oxidized to S-hydroxybutyric acid using a "1,3-butanediol dehydrogenase" that is higher in activity in the liver than in the diaphragm. M oreover the (+)-isomer of BD is oxidized at a faster rate than the (-)-isomer, suggesting that the two isomers are oxidized by two different pathways. Thesis: Sc. D., Massachusetts Institute of Technology, Department of Nutrition and Food Science, 1971; Thesis supervised by Sanford A. Miller Vita: page 196; Includes bibliographical references (pages 182-196)

An investigation of the merits of a four-element coplanar vacuum tube when used as a modulator at carrier telephone frequencies

Wed, 01 Jan 1930 00:00:00 GMT

An investigation of the merits of a four-element coplanar vacuum tube when used as a modulator at carrier telephone frequencies Perkins, Edwin H. Thesis: M.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1930; Includes bibliographical references (leaf 115).

Micro-analysis of grinding machine cuttings

Sun, 01 Jan 1922 00:00:00 GMT

Micro-analysis of grinding machine cuttings Zurlo, J. V.; Terkelsen, E. A. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1922

Tests upon bamboo as a concrete reinforcement and a consideration of its application in construction

Tue, 01 Jan 1924 00:00:00 GMT

Tests upon bamboo as a concrete reinforcement and a consideration of its application in construction Young, Joe W.; Guo, Dianbang. Thesis: B.S., Massachusetts Institute of Technology, Department of Civil Engineering, 1924

A precision method for the determination of dew points of complex gaseous systems

Wed, 01 Jan 1936 00:00:00 GMT

A precision method for the determination of dew points of complex gaseous systems Cox, John Tatum. Thesis: M.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1936; Includes bibliographical references (leaf 43).

AbsInt-AI: Language Models for Abstract Interpretation

Thu, 01 May 2025 00:00:00 GMT

AbsInt-AI: Language Models for Abstract Interpretation Wang, Michael Static program analysis is a foundational technique in software engineering for reasoning about program behavior. Traditional static analysis algorithms model programs as logical systems with well-defined semantics, enabling strong guarantees such as never missing a bug. However, traditional analyses almost always rely on uniform, hard-coded heap abstractions. While more adaptive abstractions are possible in theory, they are rarely implemented in practice due to their complexity and fragility. This limits their precision and flexibility, especially in dynamic languages like JavaScript, where heap structures are heterogeneous and difficult to analyze statically. In this work, we introduce AbsInt-AI, a language-model-guided static analysis framework based on abstract interpretation with adaptive, per-object heap abstractions for JavaScript. This enables the analysis to leverage high-level cues, such as naming conventions and access patterns, without requiring brittle, hand-engineered heuristics. Importantly, the LM agent operates within a bounded interface and never directly manipulates program state, preserving the soundness guarantees of abstract interpretation. ABSINT-AI reduces false positives by up to 34% for bug detection compared to traditional static analysis while maintaining soundness. Our ablations show that the LM’s interactions with the analysis environment are crucial, outperforming non-agentic direct LM predictions by 25%.

Optimizing Video Streaming at Scale Across Devices, Networks, and Temporal Drift

Thu, 01 May 2025 00:00:00 GMT

Optimizing Video Streaming at Scale Across Devices, Networks, and Temporal Drift Sharma, Harsha Video-streaming platforms tune dozens of playback parameters across thousands of client devices. Our measurements from Prime Video show that device-specific tuning can enhance stream quality. Yet traditional blackbox optimization methods like Bayesian optimization become prohibitively expensive due to the large configuration space and the constant emergence of new device types. We introduce AZEEM, a scalable recommendation system leveraging few-shot prediction to rapidly identify promising configurations for new devices. The key insight behind AZEEM is that devices exhibit performance similarities that enable predictions from limited observations. Trained on offline data of device-playback configuration interactions, AZEEM efficiently narrows down the search space to a small set of configurations likely to contain optimal or near-optimal candidates. Additionally, AZEEM addresses temporal distribution shift—where the best-performing configurations change over time—by recommending a small, robust set of candidates rather than a single configuration. Evaluations using largescale real-world datasets show that AZEEM reduces exploration cost by 5.8 − 13.6× and improves stream quality compared to state-of-the-art Bayesian optimization and multi-armed bandit approaches, enabling effective device-specific optimization at scale. The material in this thesis is primarily sourced from the paper "Predict, Prune, Play: Efficient Video Playback Optimization Under Device Diversity and Drift" authored by Harsha Sharma, Pouya Hamadanian, Arash Nasr-Esfahany, Zahaib Akhtar, Mohammad Alizadeh, which is currently under submission.

Oreo: Protecting ASLR Against Microarchitectural Attacks

Thu, 01 May 2025 00:00:00 GMT

Oreo: Protecting ASLR Against Microarchitectural Attacks Song, Shixin Address Space Layout Randomization (ASLR) is one of the most prominently deployed mitigations against memory corruption attacks. ASLR randomly shuffles program virtual addresses to prevent attackers from knowing the location of program contents in memory. Microarchitectural side channels have been shown to defeat ASLR through various hardware mechanisms. We systematically analyze existing microarchitectural attacks and identify multiple leakage paths. Given the vast attack surface exposed by ASLR, it is challenging to effectively prevent leaking the ASLR secret against microarchitectural attacks. Motivated by this, we present Oreo, a software-hardware co-design mitigation that strengthens ASLR against these attacks. Oreo uses a new memory mapping interface to remove secret randomized bits in virtual addresses before translating them to their corresponding physical addresses. This extra step hides randomized virtual addresses from microarchitecture structures, preventing side channels from leaking ASLR secrets. Oreo is transparent to user programs and incurs low overhead. We prototyped and evaluated our design on Linux using the hardware simulator gem5.

On Counting Substructures with Graph Neural Networks

Thu, 01 May 2025 00:00:00 GMT

On Counting Substructures with Graph Neural Networks Tahmasebi, Behrooz To achieve a graph representation, most Graph Neural Networks (GNNs) follow two steps: first, each graph is decomposed into a number of subgraphs (which we call the recursion step), and then the collection of subgraphs is encoded by several iterative pooling steps. While recently proposed higher-order networks show a remarkable increase in the expressive power through a single recursion on larger neighborhoods followed by iterative pooling, the power of deeper recursion in GNNs without any iterative pooling is still not fully understood. To make it concrete, we consider a pure recursion-based GNN which we call Recursive Neighborhood Pooling GNN (RNPGNN). The expressive power of an RNP-GNN and its computational cost quantifies the power of (pure) recursion for a graph representation network. We quantify the power by means of counting substructures, which is one main limitation of the Message Passing graph Neural Networks (MPNNs), and show how RNP-GNN can exploit the sparsity of the underlying graph to achieve low-cost powerful representations. We also compare the recent lower bounds on the time complexity and show how recursion-based networks are near optimal.

Design and Control of a Multi-Fingered Soft-Rigid Hybrid Robotic Hand

Thu, 01 May 2025 00:00:00 GMT

Design and Control of a Multi-Fingered Soft-Rigid Hybrid Robotic Hand Norton, Wil J. In robot hands, compliance improves the quality of grasps and allows for robustness in contact with the environment, which is why soft robot hands, which are inherently compliant, generate such interest despite being complex to control and model. In prior work, our lab developed a soft-rigid hybrid architecture for a robot finger, with the intention of making a compliant finger that is as easy to control as a rigid robot. This thesis details the work done to take this architecture and develop it into a five-fingered dexterous gripper capable of highly compliant grasping — over several iterations, we create an integrated tendon-driven hand that is robust, maintainable, and inexpensive. We develop a precise controller for the soft-rigid hybrid finger, and extend it for both position and task space control of the hand — additionally we implement variable stiffness control within the controller without the need for additional hardware, via adjusting gain values in the control loop. We test the ability of the hand to complete the full set of human grasping postures, and demonstrate that the soft-rigid architecture enables a high degree of generalization, able to complete 28 of the 33 identified human grasp postures. Additionally, tests illustrate the hand’s advantages in completing traditionally difficult manipulation tasks such as picking up thin deformable objects (such as a dollar bill or folding cloth) as well as in interfacing with soft or delicate target objects. We adapt a teleoperation system to map the movements of the robot gripper to a glove worn by a human operator, and evaluate the usability of the hand as a teleoperation target for completing several tasks — we illustrate promising results that the compliance of the hand compensates for operator error and allows for fast completion of tasks requiring environmental or object contact, traditionally difficult tasks for existing rigid robots. Finally, we discuss the use of the teleoperation system to record demonstrations which we then use to train an imitation learning model, utilizing an implementation of denoising diffusion probabilistic models, to complete grasping tasks. We show that our soft-rigid fingers allow a dexterous hand to be trained to perform autonomous grasping with a relatively small set of expert demonstrations, and that the compliance of the physical structure allows for variance in the environment and object position to be compensated for by the physical properties of the hand.

Development of Multi-Modality Imaging Cart for Barrett’s Esophagus

Thu, 01 May 2025 00:00:00 GMT

Development of Multi-Modality Imaging Cart for Barrett’s Esophagus Qu, Ashley Barrett’s Esophagus (BE) is a key precursor to esophageal adenocarcinoma (EAC), but current screening and risk assessment methods are ineffective and costly. Many BE cases remain undiagnosed due to asymptomatic patients, and existing risk algorithms rely on patient data rather than biomarkers. This work aims to start building a risk progression model by using a multi-modal imaging system combining autofluorescence spectroscopy, optical coherence tomography, and diffuse reflectance spectroscopy to perform label-free optical biopsies on ex-vivo tissue. These images will be co-registered and validated with histological biomarkers for BE. The ultimate goal is to develop a non-invasive endoscopic capsule and algorithm to better assess BE progression and enhance early detection of EAC.

Complexity of Basis-Restricted Local Hamiltonians

Thu, 01 May 2025 00:00:00 GMT

Complexity of Basis-Restricted Local Hamiltonians Ma, Henry A major goal of quantum complexity theory is to understand which computational problems can be solved with access to certain quantum resources. The subfield of Hamiltonian complexity specifically considers computational problems that ask about properties of local Hamiltonians, which are of critical importance in quantum complexity because they can be viewed as quantum generalizations of classical constraint satisfaction problems. In this work, we study the complexity of certain restricted variants of the Quantum-k-Sat problem, a quantum analog of the NP-complete k-Sat problem. We introduce new variants of Quantum-k-Sat which place a basis restriction on the input Hamiltonian H = Σᵢ hᵢ . Each variant is defined by a fixed collection of bases B₁, . . . , Bᵣ of n-qubit space. We require that each Hamiltonian term hi must be diagonal in one of these bases. Our results resolve the complexity of certaim basis-restricted variants of Quantum-k-Sat. First we show the Quantum-6-Sat problem with Hamiltonian terms restricted to be diagonal in an X/Z mixed basis is QMA₁-complete. Second, we combine basis restriction with the restriction of commutativity, and show the following easiness result, which applies generally to higher-level quantum systems (qudits) and bases Q and R (which are real-valued and satisfy an overlap condition): The commmuting Quantum-Sat problem on qudits, where Hamiltonian terms are either diagonal in the Q basis, the R basis, or a single mixed Q/R basis, is in NP.

Future of Personalized, Aligned Language Models

Thu, 01 May 2025 00:00:00 GMT

Future of Personalized, Aligned Language Models Han, Seungwook Aligning Large Language Models (LLMs) to cater to different human preferences, learning new skills, and unlearning harmful behavior is an important problem. Search-based methods, such as Best-of-N or Monte-Carlo Tree Search, are effective, but impractical for LLM adaptation due to their high inference cost. On the other hand, using Reinforcement Learning (RL) for adaptation is computationally efficient, but performs worse due to the optimization challenges in co-training the value function and the policy. We present a new framework for reward optimization, Value Augmented Sampling (VAS), that can maximize different reward functions using data sampled from only the initial, frozen LLM. VAS solves for the optimal reward-maximizing policy without co-training the policy and the value function, making the optimization stable, outperforming established baselines, such as PPO and DPO, on standard benchmarks, and achieving comparable results to Best-of-128 with lower inference cost. Unlike existing RL methods that require changing the weights of the LLM, VAS does not require access to the weights of the pre-trained LLM. Thus, it can even adapt LLMs (e.g., ChatGPT), which are available only as APIs. In addition, our algorithm unlocks the new capability of composing several rewards and controlling the extent of each one during deployment time. By bringing together stability, flexibility, and efficiency, we explore the future of aligned, personalized language models that can be adapted seamlessly to meet a wide spectrum of human preferences.

Post-Carbon Seoul: Low-Carbon Interventions for a High-Carbon Housing Stock

Thu, 01 May 2025 00:00:00 GMT

Post-Carbon Seoul: Low-Carbon Interventions for a High-Carbon Housing Stock Ji, Yewon Seoul, South Korea, exhibits an exceptionally rapid residential demolition-reconstruction cycle of approximately 30 - 40 years, resulting in one of the world’s shortest apartment building lifespans. This entrenched status quo, fueled by post-war policies, real estate speculation, and finance models treating housing primarily as a short-term asset, contrasts sharply with other developed nations. This research critiques South Korea’s model of rapid demolition for its significant, often overlooked, environmental impacts and social costs. To evaluate alternatives, the methodology comprises three key stages: A) a comparative analysis of the financial frameworks and sustainability outcomes characterizing Western residential longevity versus the unique Korean housing model; B) the formulation of a novel alternative practice focused on adaptive reuse and retrofitting, specifically tailored to integrate within South Korea’s economic system and cultural context; and C) the practical demonstration and assessment of this practice through a design case study, incorporating strategies like phased interventions and low-carbon materials such as mass timber. The analysis reveals that this alternative extends building lifespan and achieves substantial carbon reductions by preserving the embodied carbon within existing structures. It offers long-term financial benefits, presenting a viable economic pathway aligning key stakeholder interests through enduring value over speculative gains.

Scaling Automatic Question Generation to Large Documents: A Concept-Driven Approach

Thu, 01 May 2025 00:00:00 GMT

Scaling Automatic Question Generation to Large Documents: A Concept-Driven Approach Noorbakhsh, Kimia Assessing and enhancing human learning through question-answering is vital, especially when dealing with large documents, yet automating this process remains challenging. While large language models (LLMs) excel at summarization and answering queries, their ability to generate meaningful questions from lengthy texts remains underexplored. We propose Savaal, a scalable question-generation system with three objectives: (i) scalability, enabling question-generation from hundreds of pages of text (ii) depth of understanding, producing questions beyond factual recall to test conceptual reasoning, and (iii) domainindependence, automatically generating questions across diverse knowledge areas. Instead of providing an LLM with large documents as context, Savaal improves results with a threestage processing pipeline. Our evaluation with 76 human experts on 71 papers and PhD dissertations shows that Savaal generates questions that better test depth of understanding by 6.5× for dissertations and 1.5× for papers compared to a direct-prompting LLM baseline. Notably, as document length increases, Savaal’s advantages in higher question quality and lower cost become more pronounced.

New approaches to diagnostic imaging: Magnetic particle imaging for human functional neuroimaging and short mid-field MRI magnet design

Thu, 01 May 2025 00:00:00 GMT

New approaches to diagnostic imaging: Magnetic particle imaging for human functional neuroimaging and short mid-field MRI magnet design Barksdale, Alex Christopher Part I: Magnetic Particle Imaging for Human Functional Neuroimaging While Magnetic Resonance Imaging (MRI) has revolutionized diagnostic imaging since its clinical introduction in the 1980s — primarily focusing on hydrogen nuclei — it remains fundamentally limited by the weak nature of nuclear spin magnetism. For example, functional MRI (fMRI) provides valuable insights into brain activity through BOLD signaling, but its limited sensitivity and reliance on indirect physiological measures often necessitate large subject pools for meaningful analysis. In contrast, Magnetic Particle Imaging (MPI) utilizes the much stronger magnetism associated with superparamagnetic iron oxide nanoparticles (SPIONs), and by minimizing background signal levels which are not modulated by functional activity, it offers a promising alternative. However, there are no approved SPION tracers for human use that are well-suited to MPI, and we have little experience scaling this technology up to human-sized imagers. This thesis therefore demonstrates a human-scale MPI scanner using functional MPI (fMPI) in non-human primates and assesses its potential for future human studies. Additionally, we investigate safety aspects of MPI, specifically focusing on peripheral nerve stimulation (PNS) induced by the 25 kHz magnetic excitation fields used in MPI. Because this is a higher frequency than those used by MRI gradients, threshold data at this frequency are lacking. This thesis measures the PNS stimulation threshold in human subjects to better understand high-frequency magnetic PNS and ensure the safe implementation of human-scale MPI for future neuroimaging applications. Part II: Short Mid-Field MRI Magnet Designs Anxiety induced by the long, narrow tube of conventional 1.5T and 3T scanners is a common cause of incomplete patient examinations, leading to delays in diagnosis and reduced facility throughput. In contrast, the short aspect ratio of CT scanner bores is known to alleviate this anxiety, eliminating this problem. This thesis also addresses the need for a more patient-friendly MRI scanning option by introducing a new “hybrid” superconducting and permanent magnet concept applicable to mid-field (0.5T) superconducting solenoid magnets. While mid-field scanners offer lower sensitivity than high-field alternatives, recent advances in image reconstruction and denoising have significantly enhanced their utility, allowing them to deliver diagnostic information comparable to that of the previous generation of 1.5T scanners. Additionally, they increase the range of compatible metallic implants and offer hospitals a lower-cost, easier-to-site alternative to 1.5T and 3T scanners. They can also enhance patient comfort through shorter bore lengths and larger diameters, but their optimized winding designs still reach a limit in how short they can be made for a given homogeneity and diameter specification. This thesis introduces the use of rare-earth permanent magnets to enable further reductions in scanner length, aiming to match the aspect ratio of CT scanners.

Ab initio modeling of superconducting nanowire single-photon detectors

Thu, 01 May 2025 00:00:00 GMT

Ab initio modeling of superconducting nanowire single-photon detectors Simon, Alejandro Single-photon detectors are widely used in modern communication, sensing, and computing technology. Among these detectors, superconducting nanowire single-photon detectors (SNSPDs) possess the highest detection efficiencies, the shortest timing jitter, and the lowest dark count rates. However, for several applications, including those in the biological, astronomical, and quantum computation fields, there remains a desire to push the capabilities of modern detectors even further. To realize these improvements, it is necessary to develop an understanding of the physical mechanisms underpinning single-photon detection in these devices. However, current models are phenomenological, requiring experimental data for input, or can only recover qualitative agreement, severely limiting their predictive ability. In this thesis, we begin by describing the existing theoretical frameworks used to model superconducting materials and devices, both in equilibrium and nonequilibrium. We then illustrate an example of a phenomenological approach to modeling superconducting devices by developing an electrothermal model for the superconducting nanowire cryotron and demonstrating its efficacy in predicting the DC behavior and power dissipation of the device. Finally, we expand upon the current state-of-the-art SNSPD theory by utilizing recent advances in density functional theory to develop an ab initio model for the photon detection mechanism of SNSPDs. We then validate the predictions of our model with experimental data from the literature. The resulting model requires no experimental input, provides quantitative predictions of SNSPD performance, and can be extended to describe other superconducting devices, thus enabling the possibility of conducting a systematic search of materials for enhanced device performance.

Inference-Time Learning Algorithms of Language Models

Thu, 01 May 2025 00:00:00 GMT

Inference-Time Learning Algorithms of Language Models Akyurek, Ekin Modern language models (LMs) can perform complex tasks through in-context learning (ICL)—they can adapt to a task via examples provided in their input without any parameter updates. However, fundamental questions remain about when this adaptation works, what algorithms underlie it, and how to improve it. This thesis studies the mechanisms and limitations of ICL and develops better methods for test time adaptation of LMs on diverse benchmarks of language modeling and reasoning. I begin by evaluating the ICL capabilities of pre-trained LMs. I demonstrate that LMs can achieve strong compositional generalization when provided with few-shot examples. In a separate analysis, I show that their performance deteriorates significantly when faced with counterfactual variants of tasks they normally performed well on. Later, I develop "model problems" of ICL test the ability of LMs to learn novel mathematical structures in-context like linear functions and probabilistic formal languages. I interpret the algorithmic foundations of ICL. First, I prove that Transformer models with sufficient capacity can execute both iterative and closed-form solutions to linear regression problems, and demonstrate that these theoretical solutions manifest as interpretable intermediate variables. Then, I reveal how LMs develop specialized circuits that implement approximate n-gram learning algorithms for probabilistic languages. Building on these insights, I develop two approaches to enhance LMs. First, I demonstrate that explicitly incorporating n-gram computation into model architectures improves performance across multiple domains. Second, I introduce a test-time training that enables rapid adaptation through gradient updates on input data, achieving significant improvements over standard few-shot learning on abstract reasoning tasks. Together, these results advance our understanding of how LMs adapt to novel tasks and provide practical techniques for enhancing their test-time learning capabilities.

Trapping and Laser Cooling an Ensemble of Ytterbium-171 Atoms for use in an Atomic Clock

Thu, 01 May 2025 00:00:00 GMT

Trapping and Laser Cooling an Ensemble of Ytterbium-171 Atoms for use in an Atomic Clock Velez, Gustavo A. Optical lattice clocks require careful preparation of atomic ensembles in order to ensure homogeneous interactions with the clock laser. We demonstrate loading and laser cooling of an ensemble of ytterbium-171 atoms in a 2D optical dipole trap created by an optical cavity. Our loading method ensures that all atoms are located in the intersection of 2 perpendicular dipole traps as verified through absorption imaging. Raman sideband cooling was used to cool the atomic ensemble from 15.7 uK to 6.3 uK as measured through optical sideband spectroscopy on the 578 nm clock transition. Together, these steps improved the transfer of atoms during a Rabi oscillation from the ground to the clock state from approximately 45 percent excitation fraction to 80 percent excitation fraction. The final atomic ensemble preparation is now sufficient for running an atomic clock.

Improving and Analyzing Model Merging Methods for Adaptation

Thu, 01 May 2025 00:00:00 GMT

Improving and Analyzing Model Merging Methods for Adaptation Pari, Jyothish In this work, we explore the limitations of combining models by averaging intermediate features, referred to as model merging, and propose a new direction for achieving collective model intelligence through what we call compatible specialization. Current methods for model merging, such as parameter and feature averaging, struggle to effectively combine specialized models due to representational divergence during fine-tuning. As models specialize to their individual domains, their internal feature representations become increasingly incompatible, leading to poor performance when attempting to merge them for new tasks. We analyze this phenomenon using centered kernel alignment (CKA) and show that as models specialize, the similarity in their feature space structure diminishes, hindering their capacity for collective use. To address these challenges, we investigate routing-based merging strategies, which offer more flexible methods for combining specialized models by dynamically routing across different layers. This allows us to improve on existing methods by combining features from multiple layers rather than relying on fixed, layer-wise combinations. However, we find that these approaches still face limitations when layers within models are representationally incompatible. Our findings highlight the importance of designing new approaches for model merging that operate on well-defined input and output spaces, similar to how humans communicate through language rather than intermediate neural activations.

Evaluating Differences in GPT-4 Treatment by Gender in Healthcare Applications

Thu, 01 May 2025 00:00:00 GMT

Evaluating Differences in GPT-4 Treatment by Gender in Healthcare Applications Pan, Eileen LLMs already permeate medical settings, supporting patient messaging, medical scribing, and chatbots. While prior work has examined bias in medical LLMs, few studies focus on realistic use cases or analyze the source of the bias. To assess whether medical LLMs exhibit differential performance by gender, we audit their responses and investigate whether the disparities stem from implicit or explicit gender cues. We conduct a large-scale human evaluation of GPT-4 responses to medical questions, including counterfactual gender pairs for each question. Our findings reveal differential treatment based on the original patient gender. Specifically, responses for women more often recommend supportive resources, while those for men advise emergency care. Additionally, LLMs tend to downplay medical urgency for female patients and escalate it for male patients. Given rising interest in “LLM-as-a-judge” approaches, we also evaluate whether LLMs can serve as a proxy for human annotators in identifying disparities. We find that LLM-generated annotations diverge from human assessments in heterogeneous ways, particularly regarding error detection and relative urgency.

Highly Integrated Graphene-Based Chemical Sensing Platform for Structural Monitoring Applications

Thu, 01 May 2025 00:00:00 GMT

Highly Integrated Graphene-Based Chemical Sensing Platform for Structural Monitoring Applications López Ángeles, Christian Emmanuel Two-dimensional materials, such as graphene, hold promise for sensing applications. Graphene's remarkable surface-to-volume ratio, when employed as a transducer, enables the sensor channel to be readily modulated in response to chemical changes in proximity to its surface, effectively converting chemical signals into the electrical domain. However, their utilization has been constrained due to variations in device-to-device performance arising from synthesis and fabrication processes. To address this challenge, we employ Graphene Field Effect Transistors (GFETs) in developing a robust and multiplexed chemical sensing platform. This platform comprises a silicon chip with multiple arrays of sensing units distributed on its surface. This chip is coupled with custom-designed high-speed readout electronics for structural monitoring applications. For example, in harsh environmental conditions, structures constructed from reinforced concrete may experience degradation due to corrosion, a chemical process initiated by carbonation from atmospheric CO₂ and significant fluctuations in temperature and humidity. Under normal conditions, concrete maintains a pH level within the alkaline range of 13 to 14. However, when subjected to carbonation, its pH decreases to values between 8 and 9. Our platform excels in real-time pH monitoring. By conducting I-V sweep measurements in the sensor channel, we have established a correlation between [H⁺] concentration and the device transfer characteristics, i.e. gate-source voltage (𝑉_𝐺𝑆) at graphene's Dirac point with an accuracy of roughly 97%. Additionally, we evaluate changes in graphene channel resistance induced by pH variations. This system and correlation allow for the prompt detection of any deviations induced by corrosion within a concrete environment.

Towards More Interpretable AI With Sparse Autoencoders

Thu, 01 May 2025 00:00:00 GMT

Towards More Interpretable AI With Sparse Autoencoders Engels, Joshua While large language models demonstrate remarkable capabilities across diverse domains, the specific representations and algorithms they learn remain largely unknown. The quest to understand these mechanisms holds dual significance: scientifically, it represents a fundamental inquiry into the principles underlying intelligence, while practically–and with growing urgency– it is vital for mitigating risks from these very same increasingly powerful systems. The initial section of this thesis tackles this challenge of interpreting internal language model representations (features) by employing sparse autoencoders (SAEs). An SAE decomposes neural network hidden states into a potentially more interpretable basis. In Chapter 2, we introduce an unsupervised, SAE-based methodology that successfully identifies inherently multi-dimensional features. Notably, we establish that language models causally represent concepts such as days of the week and months of the year using circular structures. This work provided the first definitive evidence of causal, multi-dimensional features, thereby refuting the one-dimensional linear representation hypothesis. Chapter 3 further assesses whether SAEs identify “true” atomic language model features. We compare the generalization performance and data efficiency of linear probes trained on SAE latents against those trained on the original hidden state basis. The negative outcomes of these experiments suggest limitations in SAEs for capturing the true ontology of language models. Motivated by the aforementioned limitations, the second part of this thesis investigates sparse autoencoders themselves, exploring potential improvements and characterizing their failure modes. Chapter 4 examines the portion of activations not reconstructed by SAEs, which we term “Dark Matter.” We find that a significant fraction of this dark matter is linearly predictable, and furthermore, that specific tokens poorly reconstructed by SAEs remain largely consistent across SAE sizes and sparsities. This suggests that SAEs may systematically fail to capture certain input subspaces, which we hypothesize to contain inherently dense features. Subsequently, Chapter 5 investigates a method to enhance SAE utility: freezing the learned SAE parameters and finetuning the surrounding language model components to minimize KL divergence with the original model’s output distribution. This technique results in a 30% to 55% decrease in the cross-entropy loss gap incurred by inserting the SAE into the model.

Transmission Line Dynamics Modeling For Power Electronics-Enabled Control in the Electric Power Systems

Thu, 01 May 2025 00:00:00 GMT

Transmission Line Dynamics Modeling For Power Electronics-Enabled Control in the Electric Power Systems Lawson, Riley E. In the analysis and operation of electric power systems, understanding the rates at which dynamic phenomena evolve is critical. Classically, power systems operate on multiple time scales, with slower mechanical dynamics from synchronous machines, faster electromechanical controls and protection, and very fast electrical dynamics from transmission networks. This time scale separation results in system modeling techniques which neglect certain component dynamics. However, in systems with significant penetration of power electronic devices and under fast time scale phenomena, the rates at which dynamics evolve become less separated, necessitating the modeling of all system dynamics. In large-scale systems, this becomes computationally challenging due to the high dimensionality of the interconnected system model. This work investigates the role transmission line dynamics play at very fast time scales in power systems. Theoretical results are presented to analyze which transmission line dynamics contribute significantly to power system dynamics, allowing for the intelligent incorporation of transmission line dynamics into computationally tractable models. For the first time, the use of control co-design techniques are demonstrated algorithmically to design fast power electronics-enabled control to stabilize unstable dynamics in electric power systems. This technique allows the design of controls, in an iterative way, to create stable interconnected systems. Finally, transmission line modeling impacts on the design of protection on fast time scales is analyzed. This work presents techniques to protect from short circuits in response to load disconnections, and introduces DC circuit breaker configurations to cause current commutation. In the modern day, power systems operators possess the technology to implement fast control of dynamics, however, due to insufficient information on how to model and prepare for them, system operators instead rely on using conventional, overly conservative control schemes. This work aims to bridge this gap by presenting methodologies to incorporate these dynamics into next-generation system models, and how to design control and protection to mitigate the risks these fast dynamics pose.

Foundation Models for Protein Phenotype Prediction

Thu, 01 May 2025 00:00:00 GMT

Foundation Models for Protein Phenotype Prediction Calef, Robert Understanding the roles of human proteins remains a major challenge, with approximately 20% of human proteins lacking known functions and more than 40% missing context-specific functional insights. Even well-annotated proteins are often poorly characterized in diverse biological contexts, disease states, and perturbations. We present ProCyon, a foundation model for modeling, generating, and predicting protein phenotypes across five interrelated knowledge domains: molecular functions, therapeutic mechanisms, disease associations, functional protein domains, and molecular interactions. To support this, we created ProCyon-Instruct, a dataset of 33 million protein phenotype instructions, representing a comprehensive resource for multiscale protein phenotypes. By co-training a large language model with multimodal molecular encoders, ProCyon integrates phenotypic and protein data. A novel architecture and instruction tuning strategy allow ProCyon to process arbitrarily interleaved proteinand-phenotype inputs, achieve zero-shot task transfer, and generate free-form text phenotypes interleaved with retrieved protein sequence, structure, and drug modalities in a single unified model.

Functionalization of CNFET arrays for chemical sensing

Thu, 01 May 2025 00:00:00 GMT

Functionalization of CNFET arrays for chemical sensing Song, Jaekang Practical deployment of gas sensors for general-purpose applications requires integrated chips that operate at room temperature. However, real-world implementation has been limited by challenges such as the integration of highly sensitive and selective sensors, as well as insufficient statistical validation. In this work, we present an integrated gas sensor array comprising 2048 carbon nanotube field-effect transistors (CNFETs), functionalized with conductive metal-organic frameworks (cMOFs) and metal nanoparticles. Our functionalization approach enhances sensor responses by up to two orders of magnitude and enables on-chip pattern generation. Furthermore, the large number of redundant sensors allows for statistically significant measurements. The improved sensitivity is attributed to increased Schottky barrier modulation. We also demonstrate the chip’s capability to classify bacteria and yeast based on the gas mixtures emitted from cultures grown on agar plates. This work highlights the potential of integrated gas sensors as a practical, rapid, and cost-effective approach for general gas sensing applications, including biomedical applications.

Machine Learning Methods for Single Cell RNA-Sequencing Data to Improve Clinical Oncology

Thu, 01 May 2025 00:00:00 GMT

Machine Learning Methods for Single Cell RNA-Sequencing Data to Improve Clinical Oncology Boiarsky, Rebecca Single-cell RNA sequencing (scRNA-seq) offers a detailed view of the cellular and phenotypic composition of healthy and diseased tissues. While machine learning (ML) methods are well-suited for the high-dimensional nature of scRNA-seq data, current computational tools face limitations, particularly when confronted with data from clinical oncology. This thesis presents the development and application of ML techniques for scRNA-seq data to address key computational challenges, with a focus on challenges in clinical oncology. It covers four key areas: identifying gene signatures and biomarkers in multiple myeloma, developing methods to account for somatic copy number variations in tumor samples, benchmarking large, pre-trained scRNA-seq foundation models, and creating a framework for predicting clinical outcomes using patient-level representations of single-cell data. Together, these studies aim to develop and evaluate novel ML algorithms for scRNA-seq data which can unlock actionable insights for personalized medicine.

High-efficiency, low-loss Floquet Josephson Traveling Wave Parametric Amplifier

Thu, 01 May 2025 00:00:00 GMT

High-efficiency, low-loss Floquet Josephson Traveling Wave Parametric Amplifier Wang, Jennifer Advancing error-corrected quantum computing and fundamental science necessitates quantum-limited amplifiers with near-ideal quantum efficiency and multiplexing capability. However, existing solutions achieve one at the expense of the other; for example, Josephson traveling wave parametric amplifiers (JTWPAs) are highgain, broadband, and chip-based quantum amplifiers that conventionally incur a bandwidth-noise tradeoff. When operated at 20-dB gain and instantaneous bandwidths of a few GHz, JTWPAs typically reach near-quantum limited intrinsic efficiencies of 70% - 85% relative to that of an ideal phase-preserving quantum amplifier. This is due to information leakage to the sidebands of the JTWPA, which can be recovered by adiabatically transforming the input modes to Floquet modes of the system within the device. In this thesis, we experimentally demonstrate the first Floquet-mode travelingwave parametric amplifier (Floquet TWPA). Fabricated in a superconducting qubit process, this Floquet TWPA achieves minimal dissipation, quantum-limited noise performance, and broadband operation. Our device exhibits > 20-dB amplification over a 3-GHz instantaneous bandwidth, <0.5 -dB average in-band insertion loss, and the highest-reported intrinsic quantum efficiency for a TWPA of 92.1±7.6%, relative to an ideal phase-preserving amplifier. When measuring a superconducting qubit, our Floquet TWPA enables a system measurement efficiency of 65.1 ± 5.8%, the highest-reported in a superconducting qubit readout experiment utilizing phase-preserving amplifiers to the best of our knowledge. Finally, we discuss the noise limitations of our current experimental setup, as well as impedance matching strategies that will enable us to push towards ideal JTWPA performance. These general-purpose Floquet TWPAs are suitable for fast, high-fidelity multiplexed readout in large-scale quantum systems and future monolithic integration with quantum processors.

Towards Scalable Robot Learning without Physical Robots

Thu, 01 May 2025 00:00:00 GMT

Towards Scalable Robot Learning without Physical Robots Park, Younghyo The development of generalist robots—capable of performing a wide range of tasks in diverse environments—requires large-scale datasets of robot interactions. Unlike language or vision domains, where data can be passively collected at scale, robotic data collection remains costly, labor-intensive, and constrained by physical hardware. This thesis explores two complementary directions to overcome this challenge. First, we examine the limitations of training robots from scratch using reinforcement learning (RL). While RL has achieved promising results in simulation, its scalability is hindered by a largely overlooked bottleneck: environment shaping. Designing suitable rewards, action and observation spaces, and task dynamics typically requires extensive human intervention. We formalize environment shaping as a critical optimization problem and introduce tools and benchmarks to study and eventually automate this process, a necessary step toward general-purpose RL. Second, we introduce an alternative paradigm for robot data collection that does not rely on real-world robots. Using the Apple Vision Pro, we develop DART, an augmented reality (AR) teleoperation platform that streams human hand motions to cloud-hosted robot simulations. This setup enables scalable, low-latency collection of high-quality robot demonstrations without the overhead of physical setup or maintenance. Our user studies show that DART more than doubles data collection throughput while reducing operator fatigue, and policies trained in simulation using this data successfully transfer to the real world. Together, these contributions address two key bottlenecks in robot learning: the human effort required for RL environment design, and the dependence on physical robots for data. They lay the groundwork for scalable, accessible approaches to training generalist robot models.

A Reconfigurable, Distributed-Memory Accelerator for Sparse Applications

Thu, 01 May 2025 00:00:00 GMT

A Reconfigurable, Distributed-Memory Accelerator for Sparse Applications Golden, Courtney K. Iterative sparse matrix computations lie at the heart of many scientific computing and graph analytics algorithms. On conventional systems, their irregular memory accesses and low arithmetic intensity create challenging memory bandwidth bottlenecks. To overcome such bottlenecks, distributed-SRAM architectures use tiled arrays of high-bandwidth local storage to achieve very high aggregate memory bandwidth. However, current distributedSRAM architectures suffer from either poor programmability due to over-specialization or poor compute performance due to inefficient general-purpose hardware. This thesis proposes Quartz, a new architecture that uses short dataflow tasks and reconfigurable compute in a distributed-SRAM system to deliver both high performance and high programmability. Unlike traditional sparse CGRAs or on-die reconfigurable engines, Quartz allows reconfigurable compute to be highly utilized and scaled by (1) providing high memory bandwidth to each processing element and (2) introducing a task-level dataflow execution model that fits this new setting. Our execution model dynamically reconfigures tile hardware based on inter-tile messages to execute tasks on local data with fine-grained data partitioning across tiles. To make execution efficient, we explore novel data partitioning techniques that use graph and hypergraph partitioning to minimize network traffic and balance load. This is especially challenging for computations where one operand’s sparsity pattern (i.e., distribution of nonzeros) exhibits dynamic behavior across iterations, and we are the first to provide techniques to address this case. To ensure programmability, we show how a wide range of computations (expressed in an extended version of tensor algebra’s Einsum notation) and flexible data distributions can be systematically captured in small tasks for execution on Quartz. We evaluate Quartz in simulation, using an 8-chiplet design with 2,048 tiles and 824 MB of SRAM per chiplet, running six different iterative sparse applications from scientific computing and graph analytics. Quartz’s architecture, data partitioning techniques, and programming model together achieve gmean 26.2× speedup over the prior state-of-the-art programmable distributed-SRAM architecture.

Dipole Contact Engineering for Field-Effect Transistors Based on Two-Dimensional Materials

Thu, 01 May 2025 00:00:00 GMT

Dipole Contact Engineering for Field-Effect Transistors Based on Two-Dimensional Materials Gupta, Ayush Sagar In the next several years and decades, the expanded use of artificial intelligence and edge computing will demand more powerful and energy-efficient electronics. Two-dimensional (2D) semiconductors, and in particular transition metal dichalcogenides (TMDs) such as molybdenum disulfide (MoS₂), are promising candidates for future field-effect transistors. TMDs can enable aggressive lateral and vertical device scaling, and they can add computing power density and new memory and sensing capabilities via 3D integration. However, several key challenges remain before 2D-channel transistors become commercially viable, including large contact resistances at the source and drain due to the van der Waals surface of 2D materials and the Fermi level pinning effect. A variety of methods have been explored to make ohmic contacts to MoS₂, the most promising of which so far is to use semimetals such as Bi and Sb, however these materials suffer from thermal instability. This thesis addresses these challenges by (1) exploring the ultimate limit of contact metal workfunction scaling to better understand the metal-MoS₂ interface, and (2) introducing a new method of reducing contact resistance to 2D materials by inserting dipole layers at the contact interface. Initial work on ultralow-workfunction (ULWF) metal deposition on MoS₂ and subsequent device fabrication is presented, though further study is required to mitigate effects from deposition equipment and the reactive nature of these metals. In parallel, the Janus TMD MoSSe is explored as an example system for dipole contacts, with extensive material characterization of the Janus TMD MoSSe being performed, and the effect of a dipole layer on the contact properties of FETs being established. Together, these results are a significant step towards solving one of the major hurdles for the commercial introduction of 2D-channel transistors.

Specialization of Vision Representations with Personalized Synthetic Data

Thu, 01 May 2025 00:00:00 GMT

Specialization of Vision Representations with Personalized Synthetic Data Chae, Nayoung (Julia) Modern vision models excel at general purpose downstream tasks. It is unclear, however, how they may be used for personalized vision tasks, which are both fine-grained and data-scarce. Recent works have successfully applied synthetic data to general-purpose representation learning, while advances in Text-to-Image (T2I) diffusion models have enabled the generation of personalized images from just a few real examples. Here, we explore a potential connection between these ideas, and formalize the challenge of using personalized synthetic data to learn personalized representations, which encode knowledge about an object of interest and may be flexibly applied to any downstream task relating to the target object. We introduce an evaluation suite for this challenge, including reformulations of two existing datasets and a novel dataset explicitly constructed for this purpose, and propose a contrastive learning approach that makes creative use of image generators. We show that our method improves personalized representation learning for diverse downstream tasks, from recognition to segmentation, and analyze characteristics of image generation approaches that are key to this gain.

Optimizing Microservice Design Parameters

Thu, 01 May 2025 00:00:00 GMT

Optimizing Microservice Design Parameters Chen, Qihang Production-level cloud services are increasingly deployed as microservices. An important question is given application logic, how to design an effective microservice architecture. Existing studies have underscored the importance of microservice cohesiveness and coupling, using these metrics to drive automatic design optimizations. However, they have not accounted for the potential impact that such design changes may have on overall system performance, which is confirmed by our case study. In this work, we present a system that can automatically identify microservice designs that are well-balanced across performance, coupling, and cohesiveness to meet cloud provider’s requirements. the system uses a multi-round dynamic programming approach, selectively identifies promising design candidates, generates the corresponding microservice code, measures and compares the results to ultimately determine the optimal design. The designs produced by our system typically achieve over 20% throughput improvement under the same QoS with less than a 10% increase in average LCOM, and often outperform the original benchmark architectures across all evaluated metrics.

City in the River: Regeneration of the Santa Catarina River as an Intermittent Urban River

Thu, 01 May 2025 00:00:00 GMT

City in the River: Regeneration of the Santa Catarina River as an Intermittent Urban River Martínez Chapa, Daniela Full of dichotomies, the Santa Catarina River is both dry and wet, present but forgotten, central yet disconnected, valued yet feared. How should an intermittent river in a dense urban context be regenerated? This thesis reimagines its ecological, hydrological, and public potential. Set in Monterrey, Mexico, this research addresses the urgent need to rethink water management in the face of the intensifying climate crisis through different urban systems and regeneration strategies within the river basin. Focusing on the Santa Catarina River, long dismissed as a plot, void, or threat, this work proposes how an intermittent river might be re-understood not as an absence of activities or function but as a space of seasonal abundance, ecological possibility, and urban interaction. Historically engineered for control, the river has been used as a flood channel, markets, sports complexes, transportation corridors, and more. However, rarely has it been seen, treated, or protected as a river. Through the development of a pilot zone, this research suggests a replicable framework of regenerative strategies to slow down, retain, and absorb water flows, supporting both dry and wet season dynamics. These include restoring riparian ecologies, reintroducing soft edges, enabling groundwater recharge, and designing permeable, public, and accessible urban interventions that reconnect the city with the riverbed. This thesis is not a fixed proposal but a living toolkit, an adaptable model to be tested, expanded, and reimagined in the pilot as time and nature take over. At stake is not only the river’s future but also the city’s capacity to shift from resistance to relation, becoming one with it, becoming a city in the river.

Banjiha Stories (2025)

Thu, 01 May 2025 00:00:00 GMT

Banjiha Stories (2025) Park, Habin Banjiha are everywhere in Seoul. You don’t always see them—tucked below eye level, half-hidden underground—but they’re there. First built as military bunkers after the Korean War, later turned into last-resort housing, banjiha have become symbols of urban failure—spaces of neglect, flooding disasters, a problem to be erased. Both media portrayals and policy responses have advocated for their disappearance. But does removal truly protect the people who call these spaces home? This thesis moves beyond the idea that banjiha are simply failures of the city. Through three homes —three lives, it traces how these spaces are shaped, not only by policies and architecture but by the people who inhabit them. A home vulnerable to flooding, where protections exist—but not with the greatest risk. A place worn by time, held together by quiet repairs. A financial foothold in a city where affordable housing is disappearing. A space of temporary sacrifice. A shelter to return to, again and again. This is not just a story of risk or resilience, neglect or demolition. It is a story of how people live; how they adapt, negotiate, and make do in spaces that were never designed with them in mind. Rather than asking how to erase banjiha, this thesis asks: What can we learn by noticing them? What would it mean to shift the conversation—from removal to recognition, from assumption to understanding? To see these homes is to recognize not just their constraints, but the small interventions that could reshape them: a door that opens both ways so no one is trapped, policies that hold upstairs owners accountable for leaks, materials layered to prevent mold rather than mask it. Not grand reinventions, but deliberate shifts—openings for a different way forward. But before deciding what must change, we must first learn to see.

Probabilistic Inference for Inference Time Scaling of Language Models

Thu, 01 May 2025 00:00:00 GMT

Probabilistic Inference for Inference Time Scaling of Language Models Puri, Isha Large language models (LLMs) have achieved significant performance gains via scaling up model sizes and/or data. However, recent evidence suggests diminishing returns from such approaches, motivating a pivot to scaling test-time compute. Existing deterministic inference-time scaling methods, usually with reward models, cast the task as a search problem, but suffer from a key limitation: early pruning. Due to inherently imperfect reward models, promising trajectories may be discarded prematurely, leading to suboptimal performance. We propose a novel inference-time scaling approach by adapting particle-based Monte Carlo methods. Our method maintains a diverse set of candidates and robustly balances exploration and exploitation. Our empirical evaluation demonstrates that our particle filtering methods have a 4–16x better scaling rate over deterministic search counterparts on both various challenging mathematical and more general reasoning tasks. Using our approach, we show that Qwen2.5-Math-1.5B-Instruct surpasses GPT-4o accuracy in only 4 rollouts, while Qwen2.5-Math-7B-Instruct scales to o1 level accuracy in only 32 rollouts. Our work not only presents an effective method to inference-time scaling, but also connects rich literature in probabilistic inference with inference-time scaling of LLMs to develop more robust algorithms in future work. Code, videos, and further information available at probabilistic-inference-scaling.github.io/

Toward Systematic Integration of Inverter-Based Resources in Electricity Markets

Thu, 01 May 2025 00:00:00 GMT

Toward Systematic Integration of Inverter-Based Resources in Electricity Markets Pierre, Jordina This thesis introduces a multi-layer control architecture for inverter-based resources (IBRs), separating fast local feedback control from slower self-dispatch and system-level market coordination. Existing integration methods for IBRs limit their control flexibility and completely restrict their market participation potential. Two common practices include treatment of IBRs as negative loads and setting a fixed power factor during grid commissioning. Modeling IBRs as negative loads excludes them from dispatch coordination in electricity markets, significantly limiting incentive for contribution to grid reliability and flexibility. Likewise, a fixed power factor prevents the IBR from providing voltage support through reactive power absorption/injection. With a fixed power factor, constant real and reactive power limits are imposed on the inverter, even during voltage transients, ignoring the fact that an inverter’s available capacity can vary significantly due to internal current constraints and the power provided by the renewable energy source. To address the need for reactive power adjustment in IBRs and pave the way for their active participation in electricity markets , this work presents a coordinated control approach that enables IBRs to transition into active, self-dispatching participants. This thesis proposes a first layer hybrid PLL plus Q-V droop based controller in the first layer which governs millisecond-scale autonomous behavior, including low-voltage ride-through and real-time power adjustment based on voltage deviations at the point of common coupling and irradiance fluctuations from the renewable energy source, in this case solar. Given implementation from the first layer and predicted irradiance, Layer 2, which will be implemented in future work, uses a model predictive controller to provide bid functions for both real and reactive power while keeping voltage at the Point of Common Coupling within its limits. Finally, the third layer performs centralized market clearing through a security-constrained optimization by the system operator. By advocating for self-dispatched, constraint aware control, this thesis challenges the prevailing passive modeling paradigm and offers a structured, physics-informed alternative. It demonstrates how IBRs can evolve into reliable, market-integrated assets, enabling smarter renewable integration and a more resilient, cost-effective and decarbonized grid.

Approximations to worst-case data dropping: unmasking failure modes

Thu, 01 May 2025 00:00:00 GMT

Approximations to worst-case data dropping: unmasking failure modes Huang, Jenny Yijian A data analyst might worry about generalization if dropping a very small fraction of data points from a study could change its substantive conclusions. Checking this non-robustness directly poses a combinatorial optimization problem and is intractable even for simple models and moderate data sizes. Recently various authors have proposed a diverse set of approximations to detect this non-robustness. In the present work, we show that, even in a setting as simple as ordinary least squares (OLS) linear regression, many of these approximations can fail to detect (true) non-robustness in realistic data arrangements. We focus on OLS in the present work due its widespread use and since some approximations work only for OLS. Of the approximations that do not fail our tests, we find not only that a simple recursive greedy algorithm is the most conceptually straightforward but also that it can be orders of magnitude faster to run than the others.

Highly Scaled p-GaN-gate HEMTs for Low Voltage Power Electronics

Thu, 01 May 2025 00:00:00 GMT

Highly Scaled p-GaN-gate HEMTs for Low Voltage Power Electronics Darmawi-Iskandar, Patrick Rising global energy demands, driven by the advent of artificial intelligence (AI), cloud computing, and Internet of Things (IoT) devices, underscore the need for more efficient power electronics. In particular, power switches based on wide bandgap semiconductors such as gallium nitride (GaN) have emerged as promising alternatives to traditional silicon devices for low-voltage (10-100 V) applications. This work investigates the design, fabrication, and scaling of p-GaN-gate highelectron-mobility transistors (HEMTs). A p-GaN-gate epitaxial structure was developed with considerations for short channel effects. A self-aligned, gate-first process employing tungsten metallization was implemented to enable gate lengths as small as 100 nm. Device scaling was studied systematically, revealing the importance of gate aspect ratio and gate-to-drain spacing in managing short channel effects and maintaining breakdown voltage. Electrical characterization showed strong device performance, although contact resistance accounted for a substantial portion of total on-resistance. To address this, a modified fabrication approach incorporating regrown contacts was introduced, resulting in reduced contact resistance and improved overall device characteristics. The combined results highlight practical strategies for enhancing the performance and scalability of p-GaN-gate HEMTs for next-generation low-voltage power electronics.

Modelling Diarists: Diary-writing and Moral Anxieties in China, 1918–62

Thu, 01 May 2025 00:00:00 GMT

Modelling Diarists: Diary-writing and Moral Anxieties in China, 1918–62 Li, Tien Yi This thesis is a history of diary-writing in China from 1918 through 1961. Diaries are an increasingly popular but still inadequately understood primary source for historians of modern China. Previous scholars have suggested that, in the twentieth century, diary-writing became increasingly popular due to Japanese and Soviet influences, the increasing availability of manufactured blank diaries, and ruling governments that used diary-writing as a way of enforcing ideological conformity. This thesis traces an alternative history, starting from the popularization of published diaries in Shanghai in the long 1920s; to diaries’ emergence as a recognizable genre that could discoursed be theorized; to the moment the genre gained its reputation as a kind of self-expression par excellence; to its widespread inclusion into school curricula; to loosely connected attempts on the part of educators to delimit a normative way of diarywriting that, ironically, increasingly regimented self-expression. In doing so, this thesis contributes to the existing historiography by offering three correctives: I argue that 1) the initial proliferation of diaries was economically––not ideologically––motivated, 2) the popularization of diary-writing was not a concerted effort orchestrated by China’s political leaders but at best a loosely connected effort led by a middling class of educators, textbook writers, and intellectuals, and 3) diary-writing was not only regimented by communist ideology in the Maoist era but shifting moral principles and anxieties throughout the twentieth century. All in all, this thesis demonstrates the value of diaries for studying moral knowledge, epistemologies, and anxieties at the grassroots in midcentury China.

The Image of the Tunnels: Mapping Perception of the MIT Underground

Thu, 01 May 2025 00:00:00 GMT

The Image of the Tunnels: Mapping Perception of the MIT Underground Ravichandran, Shruthi Kevin Lynch’s influential book, The Image of the City, proposes five elements by which residents of a space create mental maps of their neighborhood and use these to define their spatial perception and navigation: paths, edges, districts, nodes, and landmarks. The MIT Tunnels are spaces utilized daily for a myriad of purposes: to reach labs and offices, to avoid slow-moving tourist traffic and biting Boston cold, and to explore MIT’s iconic hacking spots. This work exploresif Lynchian principles apply to these pseudourban underground spaces and culminates in a GeoGuessr-inspired virtual game where students can test and grow their knowledge of tunnel navigation. The hypotheses tested in this thesis project extend Lynch’s framework to relevant tunnel analogs: familiar paths, districts (clusters of buildings and departments), tunnel landmarks, and cross-level relationships between above- and underground mental maps. These hypotheses were tested via preliminary surveys on MIT students. Once completed, the subsequent experiments involved two games - one physically in the tunnels, one online with images of the tunnels gathered with a 360-camera. The games involved having participants navigate to a target building from a starting point. After the in-person game was completed, participants answered a series of questions about their route. These races offered information about familiar paths, landmarks, and strategies participants used to navigate the tunnels. Results from this game confirmed conclusions drawn from preliminary surveys that Lynchian principles do extend to the tunnels via relevant analogs, and above-ground knowledge and connection points offered even more information than Lynch’s five principles alone. Students consistently rely on heavily traveled paths, navigating through familiar districts, and using above ground knowledge to traverse in unknown underground buildings. This work can be extended to help grow students’ understanding of these tunnels, fostering further creativity and student expression in this complex network of spaces.

Parallel Batch-Dynamic Graph Algorithms: Coreness Decomposition and Spanners

Thu, 01 May 2025 00:00:00 GMT

Parallel Batch-Dynamic Graph Algorithms: Coreness Decomposition and Spanners Koo, Jaehyun This thesis contributes to the burgeoning field of batch-dynamic parallel algorithms by presenting parallel batch-dynamic graph algorithms for coreness decomposition and spanners, as well as a number of other related problems. The first class of problems we consider involves approximating coreness decomposition and several closely related concepts, such as (subgraph) density estimation, arboricity estimation, and low out-degree orientations. These are extremely useful structures for organizing graphs based on their density. Our algorithms process any batch of edge insertions and deletions in polylogarithmic depth while using work that is linear in the batch size (up to logarithmic factors), in the worst case. The second class of problems we consider concerns graph spanners. Over the past two to three decades, graph sparsifications that approximately preserve key graph properties have become essential tools in algorithm design. In particular, spanners—reducing the number of edges while approximately preserving pairwise distances—have been widely studied. We present the first such algorithms for computing and maintaining spanners. These algorithms achieve near-optimal amortized runtime—processing each batch in polylogarithmic depth with work nearly linear in the batch size for any number of processors.

Bridging the Sim-to-Real Gap for Athletic Loco-Manipulation

Thu, 01 May 2025 00:00:00 GMT

Bridging the Sim-to-Real Gap for Athletic Loco-Manipulation Fey, Nolan Achieving athletic loco-manipulation on robots requires moving beyond traditional tracking rewards—which simply guide the robot along a reference trajectory—to task rewards that drive truly dynamic, goal-oriented behaviors. Commands such as “throw the ball as far as you can” or “lift the weight as quickly as possible” compel the robot to exhibit the agility and power inherent in athletic performance. However, training solely with task rewards introduces two major challenges: these rewards are prone to exploitation (reward hacking), and the exploration process can lack sufficient direction. To address these issues, we propose a two-stage training pipeline. First, we introduce the Unsupervised Actuator Net (UAN), which leverages real-world data to bridge the sim-to-real gap for complex actuation mechanisms without requiring access to torque sensing. UAN mitigates reward hacking by ensuring that the learned behaviors remain robust and transferable. Second, we use a pre-training and fine-tuning strategy that leverages reference trajectories as initial hints to guide exploration. With these innovations, our robot athlete learns to lift, throw, and drag with remarkable fidelity from simulation to reality.

A Wound Designates a Subject

Thu, 01 May 2025 00:00:00 GMT

A Wound Designates a Subject Lum, Luca E. What haunts when haunting itself has been foreclosed? This thesis develops “ghostlessness” as a conceptual and aesthetic framework across my work in moving image, drawing, and writing. Ghostlessness refers to conditions that suppress haunting where it would otherwise emerge or be felt. Drawing from theoretical elaborations of hauntology, where the present is understood as structured by both suppressed pasts and unrealized futures, ghostlessness names the absence—or foreclosure—of that temporal disruption. It marks a contemporary condition in which systems oriented toward predictive governance and managed futurity preemptively neutralize rupture, sealing wounds before they can fester, reroute, or become sites of transformation. Through the works gathered here, I explore how ghostlessness functions not simply as absence but as affective and infrastructural suppression—rendering the spectral illegible, unaddressable, or unreal. Against this, my practice seeks to recapture the value of haunting in death-ridden, crisis-laden times where its presence is more prevalent than ever – hence its management, erasure, and suppression: ghostlessness.

Stylizing 3D Models With Generative AI for Fabrication

Wed, 01 May 2024 00:00:00 GMT

Stylizing 3D Models With Generative AI for Fabrication Tejedor, Leandra This thesis presents two novel approaches for modifying 3D models using generative AI for stylization while ensuring the resulting models preserve the properties required for fabrication. The first method, Style2Fab, separates functional and stylistic sections of 3D models to enable targeted modifications that preserve the model's intended functionality. By distinguishing between these sections, Style2Fab allows for alterations that maintain the model's functional purpose while providing flexibility in its aesthetic design. This approach ensures that the modified models retain their original functionality after stylistic changes. The second method, MechStyle, incorporates finite element analysis (FEA) into the generative modeling pipeline to maintain the structural integrity of the modified models. By analyzing changes in stress values during a simulated drop test at various stages of the stylization process, MechStyle restricts changes to those that preserve the model's structural viability. This ensures that the resulting models are both stylistically accurate to the user's desired results and structurally sound for 3D printing.

The Limits of Recovering Planted Subgraphs

Thu, 01 May 2025 00:00:00 GMT

The Limits of Recovering Planted Subgraphs Rajaraman, Amit Given an arbitrary subgraph H = Hₙ and p = pₙ ∈ (0, 1), the planted subgraph model is defined as follows. A statistician observes the union of the “signal,” which is a random “planted” copy H* of H, together with random noise in the form of an instance of an Erdős–Rényi graph ´ G(n, p). Their goal is to then recover the planted H* from the observed graph. Our focus in this work is to understand the minimum mean squared error (MMSE), defined in terms of recovering the edges of H*, as a function of p and H, for large n. A recent paper [MNS⁺23] characterizes the graphs for which the limiting (as n grows) MMSE curve undergoes a sharp phase transition from 0 to 1 as p increases, a behavior known as the all-or-nothing phenomenon, up to a mild density assumption on H. However, their techniques fail to describe the MMSE curves for graphs that do not display such a sharp phase transition. In this paper, we provide a formula for the limiting MMSE curve for any graph H = Hₙ, up to the same mild density assumption. This curve is expressed in terms of a variational formula over pairs of subgraphs of H, and is inspired by the celebrated subgraph expectation thresholds from probabilistic combinatorics [KK07]. Furthermore, we give a polynomial-time description of the optimizers of this variational problem. This allows one to efficiently approximately compute the MMSE curve for any dense graph H when n is large. The proof relies on a novel graph decomposition of H as well as a new minimax theorem which may be of independent interest. Our results generalize to the setting of minimax rates of recovering arbitrary monotone boolean properties planted in random noise, where the statistician observes the union of a planted minimal element A ⊆ [N] of a monotone property and a random Ber(p)^⊗N vector. In this setting, we provide a variational formula inspired by the so-called “fractional” expectation threshold [Tal10], again describing the MMSE curve (in this case up to a multiplicative constant) for large n.

Efficient Routing in the CityMesh Decentralized Fallback Wireless Network

Thu, 01 May 2025 00:00:00 GMT

Efficient Routing in the CityMesh Decentralized Fallback Wireless Network Liu, Ziqian As modern communication systems increasingly rely on centralized network infrastructure, they become more vulnerable to disruptions caused by disasters, failures, or cyberattacks. To address this risk, CityMesh proposes a decentralized fallback wireless network that leverages existing Wi-Fi devices, such as access points (APs), in buildings to maintain essential connectivity during outages. However, achieving scalable and reliable message delivery in such a network, without introducing excessive overhead, poses significant challenges. This thesis presents a new routing protocol for CityMesh, designed to operate efficiently at city scale. We first identify the limitations of traditional shortest-path source routing in CityMesh’s context, including the use of unreliable links and overhead from redundant transmissions. To address these issues, we introduce a safer path selection metric that prioritizes link reliability, a waypoint-based routing compression scheme, and a conduit mechanism to increase robustness to local failures. Our protocol further supports compact routing tables through a grid-based addressing scheme, enabling constant-size packet headers and scalable routing decisions. Additionally, we propose a suppression strategy to reduce unnecessary transmissions both between and within buildings. Finally, we extend our approach to reconnect disconnected network segments by formulating a relay placement strategy based on map data and geometric heuristics. Additionally, to reconnect fragmented network segments, we develop a practical relay placement algorithm by leveraging on the convex hull optimization and re-using global map knowledge, which ensures fast relay point computation in feasible locations such as roads and bridges. Simulations across 20 global cities show that our routing protocol achieves up to 2× higher packet delivery rates and reduces transmission overhead by up to 28× compared to GPSR under high packet loss and realistic localization error. The routing table footprint sampled across 4 randomly selected cities shows on average under 2 KB memory usage per device. Our fast relay placement algorithm also demonstrates only a small number of relays are needed to achieve full network connectivity for most of the cities, which validates CityMesh’s core premise that existing urban Wi-Fi infrastructure is sufficient to support a robust, scalable decentralized fallback network with minimal augmentation.

GPU-accelerated Inference for Discrete Probabilistic Programs

Thu, 01 May 2025 00:00:00 GMT

GPU-accelerated Inference for Discrete Probabilistic Programs Ghavami, Matin This thesis presents a comprehensive approach to GPU-accelerated inference for discrete probabilistic programs. We make two key contributions : (1) a factor graph IR implemented in JAX that supports variable elimination and Gibbs sampling, and (2) a modeling DSL with a compiler that lowers programs to the factor graph IR. Our system enables significant performance optimizations through static analysis of the factor graph structure. Variable elimination is optimized by reduction to tensor contraction with optimized contraction paths, while Gibbs sampling is automatically parallelized through graph coloring techniques. Empirical evaluations on standard benchmarks demonstrate orders of magnitude performance improvements over existing systems, with the parallelized Gibbs sampler showing speed-ups of up to 144x on Bayesian networks and even greater improvements for models with regular graph topologies such as Ising models and hidden Markov models.

The Vernaculars of Our Networks: From The Cloud to a Plurality of Grassroots Digital Infrastructures

Thu, 01 May 2025 00:00:00 GMT

The Vernaculars of Our Networks: From The Cloud to a Plurality of Grassroots Digital Infrastructures Hernandez-Cornejo, Mark A. This thesis is concerned with DIY "off-the-cloud" networks as socio-technical models that can reinscribe a community's organizational processes, identity, and culture. It questions how these networks can break away from corporate and extractive services of "the cloud" in order to achieve digital sovereignty as well as resist the hegemonic understanding of Western universal technology. Rather than grafting an outside network onto a community, how might the nodes of a network emerge from the cultural ontologies and local knowledge systems, creating a "vernacular cloud," with political, epistemic, and ontological implications? The social practice of what I call 'net/work' involves the facilitation of local digital territories that create a grassroots politics of "organic internets." In Chapter One, recent attempts to break from monopolized services like Google and Facebook are examined, providing insight into why these networks are formed and how they “de-link” from “the cloud.” Drawing from Walter Mignolo's understanding of "de-linking," the thesis argues that this process is a political project that is also epistemologically and economically non-western. Chapter Two examines the notion of 'community' in community networks through the lens of grassroots organizing such as mutual aid, delving into the care and maintenance required for system administration. Chapter Two builds on Geri Augusto's understanding of "re/trans" as a project that has developed new assemblages of knowledge and integrated them into different landscapes. It examines community networks from the Global South, where network nodes have the potential to be cosmo-ontological. Chapter Three provides examples of the principles outlined in Chapters One and Two from my work in pursuit of technical autonomy within an organization.

Sensing Buildings: Environmental Impact of Sensor Technologies and Data Infrastructure in Buildings

Thu, 01 May 2025 00:00:00 GMT

Sensing Buildings: Environmental Impact of Sensor Technologies and Data Infrastructure in Buildings Lesina-Debiasi, Simon Building operations and the construction sector are one of the largest contributors to global carbon emissions and energy consumption. While novel construction materials and insulation offer lower embodied carbon solutions, improved heating and cooling devices offer cost and energy effective building services. Above all, “smart” devices promise remote control, oversight, and optimization of building operations. With the rising implementation of AI solutions to every sector, it is important to see the digital devices as an interface to the material machinery they are connected to. The way through which we are introduced to these systems as solutions to environmental problems leaves out the operational and infrastructural costs of the devices. Making material design decisions that are conscious of the mining operations that source the rare earth minerals, to the pumping of oil for polymer coatings, to the chemical baths that separate it from the ore, all the way to the hard drives in server rigs that are cooled with water and driven by electricity, the cloud is nothing but materiality and resources. When evaluating buildings operations and construction techniques for sustainability considerations and environmental impact, connected services such as data networks and optimizations that rely on large server infrastructures and cloud computing are not part of the scope. This thesis reveals the missing components of energy evaluations in “smart” devices within the walls, floors, windows, doors, and roofs of our building, to create a framework through which building efficiency and sustainability can be reconsidered. Through historic research, literature reviews, and experiments, this work shines some light on the environmental impact of data infrastructure to which our buildings are connected. The work presented in this thesis does not claim to be comprehensive nor to solve the problem of optimizing buildings for energy efficiency. Instead, the goal is to build upon existing and established research on data infrastructure, smart technology, climate research etc. showing that, while the efforts currently taken might be improving the efficiency in a building on-site, considerations that are impacting the energy consumption off site need to be taken into consideration.

Decarbonization strategies for North American urban landscapes: Evaluating pavements and vegetation across design typologies

Thu, 01 May 2025 00:00:00 GMT

Decarbonization strategies for North American urban landscapes: Evaluating pavements and vegetation across design typologies Ramirez Cuebas, Adriana Urban landscapes are increasingly recognized as critical to climate mitigation, yet remain underrepresented in carbon accounting frameworks relative to buildings and infrastructure. This thesis advances landscape carbon assessment by introducing a typology-based Life Cycle Assessment (LCA) framework for landscape architecture. The framework integrates anthropogenic emissions and natural carbon dynamics while addressing uncertainty. It proceeds through three layers of analysis: 1) developing landscape system and project categories for carbon footprint benchmarking, 2) benchmarking the performance of the proposed landscape systems and urban typologies; and 3) assessing the mitigation potential of decarbonization strategies across systems and project types. Concrete pavers on reinforced concrete slabs and asphalt pavements (78 to 104 kgCO₂e/m²) are the most carbon intensive in the production-to-construction stage. Turfgrass and shrubs show wide variability, functioning as sources or sinks depending on species mix, maintenance, and flux magnitudes, underscoring the need for species-specific, ecologically dynamic modeling (-21 to 42 kgCO₂e/m² and -35 to 258 kgCO₂e/m²). Canopy systems act as consistent carbon sinks (-611 to -388 kgCO₂e/m² over 50 years) despite significant emissions from transportation and structural soil. Landscape systems were used to benchmark four urban typologies—streetscapes, plazas, courtyards, and urban parks. Their 50-year carbon footprints range from –80 to 21 kgCO₂e/m² in urban parks, –13 to 63 in courtyards, 22 to 79 in plazas, and 3 to 80 in streetscapes. Applying decarbonization strategies makes all typologies achieve net carbon sink status at the high bound. Urban parks achieve neutrality immediately post-construction, courtyards in 13 years, plazas in 26 years, and streetscapes by year 33. At higher emission estimates, urban parks and courtyards deepen carbon sink performance, plazas cross into net sink territory, and streetscapes approach neutrality. The detailed findings highlight the influence of planting density, maintenance regimes, and land cover composition. By structuring assessment around land covers and urban typologies, this thesis delivers a transferable carbon accounting framework aligned with design practice, offering actionable insights for embedding climate accountability into landscape architecture and public policy.

Simulation and Design of Quantum Processors for Low‑Overhead Quantum Error Correction

Thu, 01 May 2025 00:00:00 GMT

Simulation and Design of Quantum Processors for Low‑Overhead Quantum Error Correction Pahl, David This thesis investigates the simulation and design of the hardware architecture required for large‑scale quantum error correction (QEC). Specifically, we design microwave circuits for fast and high‑fidelity readout and devise a long‑range coupler (LRC) that spans five qubit lattice sites, suitable for low‑overhead quantum low‑density parity‑check (qLDPC) codes [1]. We present a prototypical nine‑qubit qLDPC code incorporating two long‑ range couplers and optimized readout circuits, achieving state‑of‑the‑art readout fidelities of up to 99.63% in 56 ns and demonstrating strong, well‑targeted couplings mediated by the LRC. Our simulations employ an efficient microwave abstraction based on ABCD transfer matrices, modeling complete qubit devices as networks of circuit elements. We use this formalism to develop a closed‑loop optimization algorithm that determines optimal readout parameters in seconds. The ABCD framework also accurately captures the multi‑mode behavior of the LRC, offering a valuable tool for developing large‑scale, low‑ overhead QEC devices.

Cost-Based Optimization for Semantic Operator Systems

Thu, 01 May 2025 00:00:00 GMT

Cost-Based Optimization for Semantic Operator Systems Russo, Matthew D. Recently, AI developers have turned to modular AI systems in order to achieve state-ofthe-art performance on challenging benchmarks and industry problems. New programming frameworks have enabled developers to build these systems by composing them out of semantic operators—i.e., LLM-powered maps, filters, joins, aggregations, etc.—inspired by relational operators from data management systems. While these systems of semantic operators can achieve strong performance on benchmarks, they can be difficult to optimize. For example, an optimizer may need to determine which model, prompting strategy, and retrieval mechanism to use for each operator. Existing optimizers are limited in the number of optimizations they can apply, and most (if not all) cannot optimize system quality, cost, or latency subject to constraint(s) on the other dimensions. In this thesis, we build an extensible, cost-based optimizer called Abacus, which searches for the best implementation of a semantic operator system given a (possibly constrained) optimization objective. The optimizer estimates operator performance by leveraging a minimal set of training examples and, if available, prior beliefs about operator performance. We evaluate the optimizer on a range of workloads including biomedical multi-label classification (BioDEX), information extraction from legal contracts (CUAD), and multi-modal question answering (MMQA). We demonstrate that systems optimized by our work achieve 18.7%-39.2% better quality and up to 23.6x lower cost and 4.2x lower latency than the next best system.

Efficient Learning and Computation of Linear Correlated Equilibrium in General Convex Games

Thu, 01 May 2025 00:00:00 GMT

Efficient Learning and Computation of Linear Correlated Equilibrium in General Convex Games Pipis, Charilaos We propose efficient no-regret learning dynamics and ellipsoid-based methods for computing linear correlated equilibria—a relaxation of correlated equilibria and a strengthening of coarse correlated equilibria—in general convex games. These are games where the number of pure strategies is potentially exponential in the natural representation of the game, such as extensive-form games. Our work identifies linear correlated equilibria as the tightest known notion of equilibrium that is computable in polynomial time and is efficiently learnable for general convex games. Our results are enabled by a generalization of the seminal framework of Gordon et al. [2008] for Φ-regret minimization, providing extensions to this framework that can be used even when the set of deviations Φ is intractable to separate/optimize over. Our polynomial-time algorithms are similarly enabled by extending the Ellipsoid-Against-Hope approach of Papadimitriou and Roughgarden [2008] and its generalization to games of non-polynomial type proposed by Farina and Pipis [2024a]. We provide an extension to these approaches when we do not have access to the separation oracles required by these works for the dual player. This work will appear in STOC 2025, [Daskalakis et al., 2025].

Explaining Black-Box Classifiers by Implicitly Learning Decision Trees

Thu, 01 May 2025 00:00:00 GMT

Explaining Black-Box Classifiers by Implicitly Learning Decision Trees Lange, Jane We present algorithms for finding two types of objects that explain the classification of a black-box model f : {±1}^d → {±1} on an instance x ∈ {±1}^d . The first is a certificate: a small set of x’s features that in conjunction essentially determines f(x). The second is a counterfactual: a nearest instance x′ for which f(x′) ≠ f(x). We obtain both algorithms via a connection to the problem of implicitly learning decision trees. The implicit nature of this learning task allows for efficient algorithms even when the complexity of f necessitates an intractably large surrogate decision tree. We solve the implicit learning task by bringing together techniques from learning theory, local computation algorithms, and complexity theory. Our approach of “explaining by implicit learning” shares elements of two previously disparate methods for post-hoc explanations, global and local explanations, and we make the case that it enjoys advantages of both. Our certification algorithm runs in time poly(d, C(f)) and outputs a certificate of size poly(C(f)), where C(f) is the “average certificate complexity" of f. Our counterfactual algorithm runs in time S(f)^[O(∆f (x))] ·log d, where S(f) is the sensitivity of f (a discrete analogue of the Lipschitz constant) and ∆f (x) is the distance from x to its nearest counterfactual. We further prove a lower bound of S(f)^[Ω(∆f (x))] + Ω(log d) for finding counterfactuals, thereby showing that the guarantees of our algorithm are essentially optimal.

Analog On-chip Training and Inference with Non-volatile Memory Devices

Thu, 01 May 2025 00:00:00 GMT

Analog On-chip Training and Inference with Non-volatile Memory Devices Lee, Jungsoo As the demand for computation in neural networks continues to rise, conventional computing resources are increasingly constrained by their limited energy efficiency. One promising solution to this challenge is analog in-memory computing (AIMC), which enables efficient matrix-vector multiplications by encoding synaptic weights into the conductance of nonvolatile memory devices. These devices are structured into crossbar arrays. To explore the potential of non-volatile memory devices in AIMC, investigations involve simulating crossbar array operations using IBM’s AIHWKIT. With this tool, I investigate the implementation of various analog computing algorithms, including TikiTaka. AIMC is evaluated for simple MNIST classification tasks and more complex deep learning models, Long Short-Term Memory (LSTM) networks. I demonstrate that devices can be categorized based on their asymmetry and non-linear weight modulation behavior. Performance improvements through the Tikitaka algorithm are observed only when the device provides a sufficient converge-dragging force; otherwise, the algorithm may even degrade performance. I also investigate how pulse-to-pulse noise and device-to-device variability affect system performance, as well as how different peripheral circuit configurations influence the overall behavior. Finally, I propose an Analog Low-Rank Adapter (Analog LoRA) by applying analog computing to the fine-tuning of large language models. I explore the necessary conditions for Analog LoRA to achieve performance comparable to its digital counterpart. Based on these findings, I present design guidelines for effectively applying analog computing to various machine learning tasks on edge devices.

CMOS-Compatible Wafer-Scale Synthesis and Rapid Characterization of Two-Dimensional Transition Metal Dichalcogenides

Thu, 01 May 2025 00:00:00 GMT

CMOS-Compatible Wafer-Scale Synthesis and Rapid Characterization of Two-Dimensional Transition Metal Dichalcogenides Jiao, Yixuan Two-dimensional transition metal dichalcogenides (TMDs) such as monolayer MoS₂ offer great promise for next generation nanoelectronics due to their atomic thickness, tunable bandgaps, and excellent electrostatic control. However, industrial semiconductor manufacturing demands CMOS-compatible, wafer-scale growth and conventional CVD methods often exceed thermal budgets and introduce contaminants, while achieving uniform, defect-free monolayers remain difficult. This thesis presents in-depth discussion on low-temperature MOCVD system design and optimization methodology for uniform monolayer TMD synthesis. We investigate the effect of alkali halide promoters (e.g. NaCl) and novel alkali-free promoters (e.g. NH4Cl and crystal violet) on synthesis of monolayer MoS₂. By optimizing the NaCl-promoted route, we achieve coalesced monolayer MoS₂ films with enlarged grain domains and demonstrate field-effect transistors with improved mobility. In parallel, we develop a CMOS-compatible crystal violet seeding method that avoids alkali metal contaminants and yields uniform monolayer coverage. To support process development, a rapid characterization pipeline was introduced: optical/SEM imaging combined with machine learning to quickly map thickness, grain size, and infer electronic quality across the wafer. These contributions collectively advance the integration of 2D TMD materials into CMOS fabrication, enabling monolithic 3D integration in future electronics.

Automating the Search for Artificial Life with Foundation Models

Thu, 01 May 2025 00:00:00 GMT

Automating the Search for Artificial Life with Foundation Models Kumar, Akarsh With the recent Nobel Prize awarded for radical advances in protein discovery, foundation models (FMs) for exploring large combinatorial spaces promise to revolutionize many scientific fields. Artificial Life (ALife) has not yet integrated FMs, thus presenting a major opportunity for the field to alleviate the historical burden of relying chiefly on manual design and trial-anderror to discover the configurations of lifelike simulations. This paper presents, for the first time, a successful realization of this opportunity using vision-language FMs. The proposed approach, called Automated Search for Artificial Life (ASAL), (1) finds simulations that produce target phenomena, (2) discovers simulations that generate temporally open-ended novelty, and (3) illuminates an entire space of interestingly diverse simulations. Because of the generality of FMs, ASAL works effectively across a diverse range of ALife substrates including Boids, Particle Life, Game of Life, Lenia, and Neural Cellular Automata. A major result highlighting the potential of this technique is the discovery of previously unseen Lenia and Boids lifeforms, as well as cellular automata that are open-ended like Conway’s Game of Life. Additionally, the use of FMs allows for the quantification of previously qualitative phenomena in a human-aligned way. This new paradigm promises to accelerate ALife research beyond what is possible through human ingenuity alone.

Uncertainty-aware Joint Physical Tracking and Prediction

Thu, 01 May 2025 00:00:00 GMT

Uncertainty-aware Joint Physical Tracking and Prediction Dasgupta, Arijit Humans possess a remarkable capacity to track and predict the motion of objects even when visual information is temporarily absent. This thesis investigates how missing sensory evidence—such as during occlusion—alters current and future beliefs about object motion, and introduces an uncertainty-aware framework to model this process. A behavioral experiment was conducted in which participants continuously predicted the future destination of a ball moving in 2.5D environments with occlusion. Results demonstrate that participants dynamically updated their predictions throughout occlusion, exhibiting adaptive belief revision and physically grounded reasoning. To model this behavior, a structured Bayesian modeling and inference approach for joint tracking and prediction was developed that integrates perception, state estimation, and future prediction in a unified process. The approach, implemented via a Sequential Monte Carlo algorithm embedded within a GPU-accelerated and parallel probabilistic programming system, maintains time-varying beliefs over both present and future object states, conditioned on observed images. These belief states are explicitly represented in symbolic form, enabling interpretable, frame-by-frame introspection of uncertainty and prediction over time. When compared against human responses, the model closely matched the temporal evolution of time-aligned decisions and outperformed plausible alternative hypotheses that failed to reason during occlusion. These findings affirm that the absence of changing visual evidence does not engender a void in physical reasoning, but is evidence in itself—processed and revised through structured, probabilistic inference. By integrating probabilistic programming with human behavioral data through structured Bayesian modeling and inference, this thesis advances a computational account of intuitive physical reasoning and provides a foundation for building interpretable, uncertainty-aware AI systems that mirror human-like physical intelligence.

Toward an Age-Ready Suburbia

Thu, 01 May 2025 00:00:00 GMT

Toward an Age-Ready Suburbia Du, Minghao; Zhuang, Kaicheng As America’s population ages, suburban neighborhoods face urgent challenges. Originally designed for young, car-dependent families, the suburban landscape today often presents barriers to aging in place, including poor walkability, inaccessible housing, and limited access to essential services and care. This thesis investigates these challenges and proposes a strategy for reimagining suburban environments through demographic analysis, spatial mapping, persona-driven research, architectural prototyping, and community planning. It traces the historical evolution of suburbia, critically evaluates existing senior housing typologies, and advances new frameworks for retrofitting residential neighborhoods to better support aging populations. Focusing on Sacramento, California, the research identifies high-priority areas where aging, affordability challenges, and mobility barriers intersect. Grounded by a pilot care home project, the study demonstrates how modest interventions, such as retrofitting single-family homes into small-scale residential care environments, can enhance both livability and care access. The first phase of the pilot project has been constructed, offering a demonstration of the proposed model’s feasibility. A phased development and financial strategy are also outlined to ensure broader applicability. While rooted in Sacramento, the thesis offers a framework relevant to many suburban contexts across the United States, particularly naturally occurring retirement communities (NORCs) where older adults are aging in place. Rather than creating isolated senior enclaves, the work promotes a distributed, community-integrated model that strengthens neighborhood resilience and supports intergenerational living. By combining design innovation with policy awareness and development feasibility, the thesis presents a scalable and adaptable approach to reshaping suburbs for an aging society.

Calibration and Control of Superconducting Qubits for Low‑Overhead Quantum Error Correction

Thu, 01 May 2025 00:00:00 GMT

Calibration and Control of Superconducting Qubits for Low‑Overhead Quantum Error Correction Pahl, Lukas The ability to coherently and reliably manipulate quantum information marks a fundamental technological leap—realizable through a universal, fault‑tolerant quantum computer. Achieving this goal requires progress across all layers of the quantum computing stack, from physical qubits to theoretical algorithms. In this work, we address multiple layers of this stack. We develop a software architecture for scalable device calibration using modular calibration graphs. We introduce real‑time frequency stabilization techniques, demonstrating improved single‑qubit gate fidelities and progress toward multiqubit feedback. Finally, we explore how quantum error correction overhead can be reduced using low‑density parity‑check codes. We present logical protocols for a non‑local nine‑qubit code, which significantly outperforms comparable surface code implementations in both qubit efficiency and computational capability. These results represent practical steps toward overcoming key challenges in fault‑tolerant quantum computing.

ModelDiff: A Framework for Comparing Learning Algorithms

Thu, 01 May 2025 00:00:00 GMT

ModelDiff: A Framework for Comparing Learning Algorithms Shah, Harshay We study the problem of (learning) algorithm comparison, where the goal is to find differences between models trained with two different learning algorithms. We begin by formalizing this goal as one of finding distinguishing feature transformations, i.e., input transformations that change the predictions of models trained with one learning algorithm but not the other. We then present ModelDiff, a method that leverages the datamodels framework (Ilyas et al., 2022) to compare learning algorithms based on how they use their training data. We demonstrate ModelDiff through three case studies, comparing models trained with/without data augmentation, with/without pre-training, and with different SGD hyperparameters. Our code is available at https://github.com/MadryLab/modeldiff.

Sanctuary for Who?

Thu, 01 May 2025 00:00:00 GMT

Sanctuary for Who? Salazar, Juan Philadelphia, often recognized as the poorest major city in the United States, became a Sanctuary City in 2014. The designation committed the region to policies limiting cooperation with federal law enforcement in the persecution of undocumented communities. Policies have ranged from refusing to detain individuals without judicial warrants to prohibiting Immigration and Customs Enforcement (ICE) from accessing municipal databases or facilities for detention purposes. At the community level, the notion of the Sanctuary City sought to promote organizing against unlawful persecution of residents. Over the past eleven years, however, the framework of protection it promised has faltered under mounting federal pressure. The Sanctuary City's symbolic authority and limited scope have failed to shield residents from persecution or restrict ICE's intensifying operations within the area. In 2019, Juntos, the city's foremost immigrant advocacy organization, criticized Philadelphia's Sanctuary status as inadequate. Cited the ongoing persecution of its communities and the declining quality of life for all residents, the organization urged the city to abandon the term "Sanctuary." They instead petition the city to focus instead on meaningfully protecting all residents of Philadelphia, stating, "Let us instead work together to build the kind of city we all want to live in." Junto's critique forms the basis of this thesis, using it as an invitation to reimagine the Sanctuary City as a shift from a policy framework toward a general ethic and design sensibility. This thesis proposes that Philadelphia's crux, like all cities, lies in its ability to sustain communities' pursuit of a dignified life. As a primary agent in the formation of cities, the architect must then make this struggle their own and deploy the tools of their discipline to protect life and inspire dignity. By framing Philadelphia as a city shaped by deindustrialization, disinvestment, and policing, the thesis explores how architecture can respond to these forces by reviving the city's industrial character and establishing new boundaries able to safeguard community rights. Integrating legal, spatial, and semantic insights from federal authorities' rules of engagement will provide novel typologies and programs for the city that address its systemic inequities while fostering environments where life and dignity can flourish. By inscribing meaningful boundaries, and re-equipping the city to make for itself, the thesis suggests architecture becomes a tool for collective protection and urban regeneration.

Structural analysis at scale: Computational modeling of embodied carbon in complex floor layouts

Thu, 01 May 2025 00:00:00 GMT

Structural analysis at scale: Computational modeling of embodied carbon in complex floor layouts Hirt, Natasha K. To meet the needs of growing populations, rates of new construction are increasing at a record pace worldwide. The built environment, already one of the single largest contributors to global CO₂e emissions, will become a significant environmental challenge in the coming decades. To mitigate the anticipated environmental impact of future construction, we need to rethink how we build. One strategy, which is the subject of this work, is improving the material efficiency of flexural systems like floors. Floors are among the most materially wasteful structural components in buildings, and while decades of research have explored optimal floor system design, the complexity of proposed solutions has limited their practical implementation. Furthermore, the industrial tools available to structural designers do not lend themselves to flexible experimentation or large-scale analysis. As a result, most flexural systems today rely on approximations and rules of thumb rather than mathematically optimal designs, data-driven decision making, or iterative design processes. This thesis bridges the gap between practical engineering, material efficiency, and design freedom. It presents novel, code-compliant tools for the computational analysis and optimization of flat slabs supported by a network, or grillage, of beams, using a model system of reinforced concrete supported by steel W-sections. The method is used to perform a large-scale analysis of 24,192 unique combinations of beam topologies and assembly design decisions. The results of this analysis find improvements in structural embodied carbon of up to 53.4% over the business-as-usual design case, and also yield generalizable takeaways about the key factors influencing material efficiency in floor slabs. One of the advantages of the method is its flexibility in taking on a range of complex design challenges. These are presented as extensions to the method, and include designing with a constrained inventory for a series of real-world case studies, and automatically deriving novel structural geometries from dense ground structures. The method and results shown in this thesis expand the range of analysis tools that engineers have access to, enabling a wide range of creative designs and explicitly linking design decisions to environmental impact.

Inscrutability: An Epistemological Experiment

Thu, 01 May 2025 00:00:00 GMT

Inscrutability: An Epistemological Experiment Huang, Brian Hudson Through four different projects, this thesis explores the idea of dimensions of representation, a concept introduced by 20th century French philosopher Michel Foucault in his book The Order of Things. Foucault argues that the Classical episteme, which Foucault defines as the discourse surrounding knowledge-making that lasted from the 17th century to the 19th century, was determined by the idea of dimensions of representations. Dimensions of representations states that during the Classical episteme, knowledge was formulated by representations of the external world, such as through systems of classification, ordering, and relations, rather than through resemblance. The first project, Holes in the Sieve (2023) will address the problematics of classification through a infamous case in the history of paleoanthropology: the Piltdown Man. The second project, Contrapposto in Space (2024) addresses how representation has been instrumentalized in technoscience through space research. Finally, the last two projects, the Poem Box (2024) and Micropoetry (2025) posit a way forward at the limits of representation by engaging with semiotic theory. By engaging with language games, poetry opens up the possibility to deny the position of being knowable, allowing one to disappear into inscrutability.

Koalisi Lahan–Gambut: Assembling Peat–Land Futures in Kalimantan

Thu, 01 May 2025 00:00:00 GMT

Koalisi Lahan–Gambut: Assembling Peat–Land Futures in Kalimantan El Haq, Haidar Throughout Indonesia’s colonial and postcolonial histories, the peatlands of Kalimantan have been not only politically contested spaces but also sites of ontological struggle. From transmigrasi programs to Suharto’s Mega-Rice Project and most notably today’s carbon offset regimes, peat has been transformed into a paradoxical ecology: degraded yet investible, conserved yet profitable. These transformations enclose land, force communities to choose between extraction or restoration, criminalize fire, and abandon regenerative forms of cultivation. These are histories of ontological occupation institutionalized: the marginalization of both peat’s inhabitants and the soil itself as world-making agents, shaped by speculative regimes of governance, rooted in planetary imaginaries of climate salvation and fantasies of productivity. This thesis proposes Koalisi Lahan–Gambut (Peat–Land Coalition), a speculative parainstitution that explores how coalitional spatial practices might reclaim inhabitation in peat ecologies. Situated in a Ngaju village within the buffer zone of one of the world’s largest carbon offset territories—between deep peat and riverine edges, between restoration enclosures and plantation areas—the coalition works through the murkiness of peat, the heterogeneity of its inhabitants, and the crowded terrain of overlapping institutional claims. It foregrounds the frictions between gambut (peat) and lahan (land). Structured across three inquiries, the document presents a Living Glossary that assembles field terms and relational epistemologies drawn from Kalimantan’s peatlands; a genealogy of Governance, Carbon Fix, and Buffer Zone that traces the historical and institutional processes that rendered peatlands governable; and Landing in the Buffer Zone, which turns to the coalition’s situated experiments in becoming-with, inhabiting, and reclaiming the space between peat and land.

Engineering TEV Protease Specificity: An Exploration of Machine Learning and High-Throughput Experimentation for Protein Design

Thu, 01 May 2025 00:00:00 GMT

Engineering TEV Protease Specificity: An Exploration of Machine Learning and High-Throughput Experimentation for Protein Design Sundar, Vikram Engineering sequence-specific proteases would enable a wide variety of therapeutic applications in diseases ranging from cancer to Parkinson’s disease. However, many previous experimental and physics-based attempts at protease engineering have failed to engineer specificity in cleaving alternative substrates, rendering them useless. In this thesis, we aim to engineer TEV (tobacco etch virus) protease, a highly sequence-specific protease, to cleave alternative substrates. We incorporate novel high-throughput assays and powerful machine learning (ML) methods for highly effective protein engineering. The first portion of this thesis focuses on generating fitness landscapes from high-throughput experiments. Most machine learning models do not account for experimental noise, harming model performance and changing model rankings in benchmarking studies. Here we develop FLIGHTED, a Bayesian method of accounting for uncertainty by generating probabilistic fitness landscapes from noisy high-throughput experiments. We demonstrate how FLIGHTED can improve model performance on two categories of experiments: single-step selection assays, such as phage display, and a novel high-throughput assay called DHARMA that ties activity to base editing. FLIGHTED can be used to generate robust, well-calibrated fitness landscapes, and when combined with DHARMA, our methods enable us to generate fitness landscapes of millions of variants. We then evaluate how to model protein fitness given a fitness dataset of millions of variants. Accounting for noise via FLIGHTED significantly improves model performance, especially of high-performing models. Data size, not model scale, is the most important factor in improving model performance. Furthermore, the choice of top model architecture matters more than the protein language model embedding. The best way to generate sufficient data scale is via error-prone PCR libraries; models trained on these landscapes achieve high accuracy. Using these methods, we successfully engineer both activity on an alternative substrate and specificity when compared to the wild-type. The ML-designed variants outperform anything found in the training set, demonstrating the value of machine learning even with experimental libraries of millions of variants. However, our results are limited to relatively close substrates. How best to improve model performance on distant substrates remains an open question.

An Experimental Study on the Effects of Three-Piece Oil Control Ring Design and Liner Finish on Lubricating Oil Consumption in a Hydrogen-Fueled Single-Cylinder Reciprocating Engine

Thu, 01 May 2025 00:00:00 GMT

An Experimental Study on the Effects of Three-Piece Oil Control Ring Design and Liner Finish on Lubricating Oil Consumption in a Hydrogen-Fueled Single-Cylinder Reciprocating Engine Tamburro, Alexandra Reducing lubricating oil consumption (LOC) in reciprocating engines is an increasingly important objective in the pursuit of lower greenhouse gas emissions, longer maintenance intervals, and compliance with tightening environmental regulations. In 2022, the U.S. transportation sector alone was responsible for 29% of national greenhouse gas emissions, 87% of which originated from systems powered by reciprocating engines [1]. While significant progress has been made in fuel efficiency, oil consumption remains as a key contributor to carbon emissions. This research investigates the impact of design parameters in three-piece oil control rings (TPOCRs) and liner surface finish on oil consumption behavior. Utilizing a hydrogen-fueled engine—where the only source of CO₂ emissions is from consumed lubricating oil—this study develops a high-fidelity, FTIR-based method for direct LOC measurement. A derivation of oil consumption based on air and fuel mass flow rates and measured CO₂ emissions is presented, alongside a sensitivity analysis which identified FTIR measurement uncertainty and ambient CO₂ variation as dominant error sources. All experiments were conducted at 2000 RPM under medium load (4 bar IMEP). The experimental results showed that under the tested condition, 1) increasing liner roughness increases the LOC and 2) changing the orientation of any rails with asymmetrical profile to favor up-scraping results in an elevation of LOC. Analyses applying liner vaporization and TPOCR models showed that the changes in liner oil film thickness brought by the TPOCR changes have negligible effect on the LOC from the oil evaporation. Increases in upper-rail up-scraping ability and the oil accumulation inside the TPOCR groove can both elevate the LOC although further investigation is needed to understand the oil transport paths leading to the LOC. This work provides a foundation for future optimization of TPOCR design by highlighting key ring-liner interactions and oil transport mechanisms. Further study of asymmetric geometries and surface characteristics will provide further insights for reducing oil consumption in engine platforms.

Design of Future Energy Infrastructure: Understanding trade-offs between Renewable Capacity, Storage and Transmission Networks for Low-Carbon Landscape

Thu, 01 May 2025 00:00:00 GMT

Design of Future Energy Infrastructure: Understanding trade-offs between Renewable Capacity, Storage and Transmission Networks for Low-Carbon Landscape Bhupathi, Hari Raghavendran In 2021, the United States committed to achieving net-zero greenhouse gas emissions by 2050, requiring a fundamental transformation of its energy infrastructure. This thesis develops a nationwide optimization model to minimize capital expenditures and understand the trade-off between renewable capacity, storage, and transmission networks. The results show that the least-cost configuration, achieved when nuclear and battery capital costs fall by 50%, requires approximately $3.25 trillion in new investment - a 37% reduction relative to the baseline scenario. Comparative scenario analysis reveals a marked shift toward centralized storage when nuclear costs decline, which improves reliability and reduces contingency requirements - mirroring inventory pooling dynamics in supply chains. Concurrently, wind capacity additions fall sharply, with each 10% reduction in nuclear cost halving the predicted wind capacity addition. Transmission infrastructure evolves accordingly: 765 kV lines decline as nuclear becomes more decentralized, while 230 kV lines expand modestly to manage increased intermittency. By quantifying trade-offs across technologies and identifying system tipping points, this work offers a framework for policymakers and long-horizon investors.

Barometer-Based Tactile Sensing: Characterization, Processing, and Applications for Dynamic Manipulation

Thu, 01 May 2025 00:00:00 GMT

Barometer-Based Tactile Sensing: Characterization, Processing, and Applications for Dynamic Manipulation Shah, Sharmi Reliable tactile feedback is essential for robotic systems to interact effectively with their environments, especially in dynamic manipulation tasks where detecting contact onset, direction, and force is critical for control and planning. This thesis advances the development of barometer-based tactile sensors for low-force interactions, building upon prior work from the Biomimetic Robotics Lab. Previous work demonstrated that neural networks could infer contact location and three-axis contact force from barometers embedded within an elastomer. However, these models did not account for the viscoelastic behavior of the elastomer, which degrades sensor repeatability and bandwidth. To address these limitations, this thesis introduces a recurrent neural network (RNN) architecture that captures viscoelastic transients in the sensor response. The proposed methods are evaluated on two sensor geometries: a spherical sensor and a slimmer ellipsoid variant. An automated data collection pipeline is developed to generate temporally-continuous, uniformly sampled datasets across the sensor surface. RNN models trained on this data show that temporal modeling improves force prediction accuracy across both designs. To improve angle prediction accuracy, a binning strategy is used to enforce a uniform prior over contact orientations. The resulting "Binned RNN" neural networks are small-scale and demonstrate high sensitivity, enabling responsive tactile feedback. The utility of these tactile sensors is demonstrated by integrating the sensors onto a dexterous two-finger gripper and performing light grasping and estimation of object reorientation using solely tactile measurements. This work shows that accounting for viscoelastic effects through informed sampling and temporal modeling enhances the practical performance of elastomer-based tactile sensors in robotic systems.

Can diffusion models capture extreme event statistics?

Thu, 01 May 2025 00:00:00 GMT

Can diffusion models capture extreme event statistics? Stamatelopoulos, Stamatios For many important problems it is essential to be able to accurately quantify the statistics of extremes for specific quantities of interest, such as extreme atmospheric weather events or ocean-related quantities. While there are many classical approaches to perform such modeling tasks, recent interest has been increasing in the usage of generative models trained on available data. Despite the sporadic success of such methods, it is not clear for what systems or datasets a system-agnostic generative AI tool is capable of generating previously ‘unseen’ extreme events in a manner that accurately extrapolates the tails for the observable of interest. Here, we propose an apriori criterion, which based on the geometry of the training dataset, it can predict whether a generative AI tool will be able to extrapolate the tails, i.e. generate previously unseen extreme events. The idea is to quantify whether existing extreme events lie in the interior of the dataset or its boundary. In the former case it is shown that generative AI tools can work in an ‘interpolation’ mode and generate new extreme events. On the other hand, if the topology of the dataset is such that extremes live in the boundary of the domain then the generative AI algorithm needs to operate in an extrapolation mode which does not lead to accurate results. We illustrate our findings on a specific class of Diffusion Models (DMs) called Denoising Diffusion Probabilistic Models (DDPMs) and we test on three datasets, a simple on-hyperball dataset following a Weibull distribution for the radii of the data points of dimensionality 2 • 10³, a dataset sampled from the so-called Majda-McLaughlin-Tabak Wave Model (MMT), of dimensionality 8.1 • 10³ and a dataset consisting of Lagrangian turbulence trajectories, of dimensionality 2 • 10³.

Face Me, I face you: Towards an Indigenous Economy of Glass in Southern Nigerian Dwellings

Thu, 01 May 2025 00:00:00 GMT

Face Me, I face you: Towards an Indigenous Economy of Glass in Southern Nigerian Dwellings Ajienka, Soala Lolia This thesis proposes the weaving together of two lost traditions - the practice of primary glassmaking in southern Nigeria and the U-shaped bungalow typology of multi-family housing - as a means to address both the qualitative and quantitative housing deficits in Port Harcourt and to support the broader requisites of macroeconomic productivity in Nigeria. The thesis frames the argument that the materiality and application of glass can reconnect the inhabitation and construction of Face Me, I Face You (FMIFY) housing to Nigerian history, culture, and identity. By charting a blueprint for localized material production and engaging questions of affordability, cost structure, and financing, this work positions design as a technical solution and an act of cultural authorship. As an architect, builder, and member of the community, I advocate for a new practice in which the bond between local craftsmanship and housing development is re-established - through material choices, construction systems, economic benchmarking and spatial design strategies. This body of work braids together three interconnected narratives: First, it traces the historical evolution of the U shaped bungalow typology, revealing its roots as a colonial adaptation of the rural compound house, the economic conditions that have led to its physical obsolescence yet sustained market relevance and examining how its cultural significance was gradually diluted through climate-insensitive design and the introduction of imported materials. Second, this body of work rediscovers Nigeria’s precolonial glassmaking traditions, with a focus on artisanal production methods that offer environmental efficiency, energy intelligence, and deep cultural resonance - qualities in stark contrast to the high-energy, standardized imported glass that dominates today’s housing. Third, it integrates these two recoveries through built interventions: redesigning roof structures to support artisanal glass rondels, optimizing daylighting, ventilation, and thermal comfort, and reorganizing courtyards to revive their role as culturally vibrant, socially essential spaces. By leveraging indigenous glassmaking practices and small-batch production models, this thesis advocates for the creation of a circular economy, generating local employment, reducing embodied energy, and restoring cultural resilience - while delivering environmentally sensitive and economically viable housing solutions that demonstrate comparable return on costs for their owners. Foregrounding opacity as a design value, the project seeks to balance communal life with cultural and spatial notions of privacy, challenging the hegemony of imported transparency. Through the strategic curation of apertures, the careful modulation of light and shadow, and the integration of locally crafted glass rondels, the thesis re-envisions the Face Me I Face You typology. Ultimately, this work positions artisanal glass not only as a building material, but as a medium for recalibrating housing production in southern Nigeria toward systemic resilience and self-determination.

DexWrist: A Robotic Wrist for Constrained and Dynamic Manipulation

Thu, 01 May 2025 00:00:00 GMT

DexWrist: A Robotic Wrist for Constrained and Dynamic Manipulation Ulloa, Gabriella E. DexWrist is a compliant robotic wrist designed to advance robotic manipulation in highly-constrained environments, enable dynamic tasks, and speed up data collection. DexWrist is designed to be close to the functional capabilities of the human wrist and achieves mechanical compliance and a greater workspace as compared to existing robotic wrist designs. The DexWrist can supercharge policy learning by (i) enabling faster teleoperation and therefore making data collection more scalable; (ii) completing tasks in fewer steps which reduces trajectory lengths and therefore can ease policy learning; (iii) DexWrist is designed to be torque transparent with easily simulateable kinematics for simulated data collection; and most importantly (iv) expands the workspace of manipulation for approaching highly cluttered scenes and tasks. More details about the wrist can be found at: https://sites.google.com/view/dexwrist/home.

Guiding Labor: Sensable Instructions through Digital Jigs

Thu, 01 May 2025 00:00:00 GMT

Guiding Labor: Sensable Instructions through Digital Jigs Griffin, Danny Contemporary architects find themselves at a juncture, navigating the transition from traditional modes of instruction to an asymmetrical integration of digital technologies. Drawings remain central to architectural practice, yet a widening gap persists between tools for making drawings and tools for interpreting them. Since Alberti’s division between intellectual and productive labor, architectural instructions have been generated in remote offices and executed on distant construction sites. Digital tools have expanded the information density of drawings, yet the process of interpretation remains predominantly analog. Graphical conventions, though precise, are abstract, and so paper instructions alone lack spatial meaning. Builders ultimately rely on the aid of analog locating techniques to translate these abstractions into actions. Tools as simple as strings and squares have long been present on construction sites, enabling this translation. Over time, the shape and function of such devices has evolved in response to different pressures of location, from the Gothic template which left room for the builder to improvise, to the industrial jig that constrained movement to ensure replicability. The limitations of analog locating became clear when the plumb bob, long trusted to mark which direction was vertical, proved inadequate for navigating trajectories of flying objects. The solution was to embed physical devices with memory, marking a transition from tools which measure where they are to those that know where they are going. This shift from stateless to stateful devices gradually entered construction sites, and though we might distrust the devices that make possible the steering of missiles, this paradigm shift offers a productive challenge to the field of architecture. If simplifying complex construction is worthwhile, then communication pathways which more faithfully transfer information from digital model to physical destination must be explored. Central to this transformation are the tools which anchor instructions on site: interfaces already mediating between architect and builder, which must now evolve to interpret digital signals from afar. Digital jigs will be the conduits of paperless instruction on physical sites, enabling what this thesis terms sensable instructions: instructions receivable by both machines and humans.

Natural Interaction: 3D Modeling in Wearable VR Using a Gesture and Speech Interface

Thu, 01 May 2025 00:00:00 GMT

Natural Interaction: 3D Modeling in Wearable VR Using a Gesture and Speech Interface Bei, Yining Designers often rely on keyboard and mouse for 3D modeling, a method that can feel unintuitive or restrictive—especially in collaborative or spatially immersive settings. This thesis explores how multimodal interaction, specifically the combination of hand gestures and voice commands, can support more natural, efficient, and accessible 3D modeling in virtual reality (VR). Built on a custom Unity-based system integrating Meta Quest hand tracking and Wit.ai voice recognition, the study investigates how these two input modes—gesture and speech—can be used together to manipulate and modify 3D geometry in real time. The research proceeds in three phases: (1) a formative study analyzing how users intuitively deploy gestures, revealing common preferences, task breakdown strategies, and limitations in gesture inputs; (2) system design and implementation of both gesture-only and gesture + speech interfaces for navigation and object manipulation (e.g., translation, scaling, duplication); and (3) a comparative user study evaluating gesture-only, gesture + speech, and keyboard + mouse workflows in terms of learning curve, task efficiency, and user satisfaction. Results show that gesture + speech enables smoother transitions across modeling subtasks and allows users to offload certain parameters (e.g., numeric values, distances) to voice while using gestures for spatial control. Participants reported higher engagement and lower cognitive load compared to keyboard-based workflows, especially in tasks involving spatial scale and collaboration. This thesis demonstrates the feasibility and design potential of multimodal interaction for immersive modeling workflows and offers insights for future XR design tools that seek to blend precision with embodied interaction.

Learning-Guided Optimization for Intelligent Mobility Systems

Thu, 01 May 2025 00:00:00 GMT

Learning-Guided Optimization for Intelligent Mobility Systems Li, Sirui Efficient and reliable mobility systems are essential to modern-day society, with broad impacts ranging from day-to-day commuting, public transportation, emergency response to last-mile package delivery and freight logistics. Autonomous vehicles have the potential to improve mobility efficiency and convenience but also raise questions about reliability and feasibility of deployment. The first contribution of this thesis is a set of novel, principled control-theoretical analyses that provide strong stability and reliability guarantees for autonomous vehicles and human-compatible driving, and they further covers emergent traffic behaviors in mixed-autonomy systems. While these theoretical guarantees offer valuable insights, mobility systems are inherently complex, and their overall performance often relies on solving difficult optimization problems, many of which are combinatorial, thus presenting significant scalability challenges. Overcoming these challenges requires innovative approaches that extend beyond traditional control techniques. This thesis further contributes a set of machine learning-guided optimization algorithms that significantly enhance the efficiency and scalability of solving combinatorial optimization problems. These algorithms have proven effective across a wide range of mobility-related applications. Compared to state-of-the-art solvers, they achieve 10× to 100× speed-up in large-scale vehicle routing problems, 35% to 70% solve-time improvement in various mixed-integer linear programming problems, and up to 54% acceleration in long-horizon scheduling problems. These advancements open new possibilities for efficient decision-making in large-scale transportation systems, enabling smarter, faster, and more adaptive mobility solutions. Combining learning, optimization, and control, this thesis demonstrates the potential of learning-guided optimization and principled control-theoretical analysis to address the increasing complexity of modern mobility systems.

Top-down and bottom-up interactions for cortical bursting

Thu, 01 May 2025 00:00:00 GMT

Top-down and bottom-up interactions for cortical bursting Tang, Vincent D. High-frequency burst firing occurs throughout the mammalian cortex in vivo, yet both the underlying mechanisms and functional roles of bursts are unclear. Burst firing in brain slices is strongly modulated by the activity of apical dendrites, which branch extensively in layer 1 (L1) and receive long-range inputs from higher-order cortical and thalamic areas. These properties suggest a powerful subcellular substrate by which single pyramidal neurons could multiplex bottom-up and top-down information via L1-independent tonic spikes and L1-dependent bursts, respectively, and have provided a basis for emerging theoretical models of cortical computation and learning. However, our understanding of burst firing and subcellular processing remains critically limited by a lack of evidence in awake animals. It is unclear whether burst firing a) is preferentially recruited by bottom-up versus top-down inputs, and b) requires apical dendritic engagement. To answer these questions, we performed high-density extracellular recordings in primary visual cortex of awake mice while presenting a battery of Gabor (bottom-up) and inverse (top-down) visual stimuli. We report widespread high-frequency bursts in L2/3 and L5 pyramidal neurons. Contrary to expectation, bursts exhibited extremely short response latencies, and were most strongly recruited by Gabor stimuli. We further tested the causal contribution(s) of apical dendrites to burst firing and top-down visual tuning via two optogenetic manipulations: direct L5 apical tuft inhibition and NDNF interneuron activation. Strikingly, L1 inhibition only modestly reduced the burst fraction, and did not differentially affect Gabor vs inverse responses. Taken together, these results challenge prevailing theories of apical dendritic involvement in burst spike generation and feedback visual tuning, and provide new biological constraints for future theoretical and experimental work.

Understanding the Limits of Coupled Condensation and Desorption in Sorption-Based Atmospheric Water Harvesting (SAWH) Devices

Thu, 01 May 2025 00:00:00 GMT

Understanding the Limits of Coupled Condensation and Desorption in Sorption-Based Atmospheric Water Harvesting (SAWH) Devices Stamler, Natasha Lia Access to clean water is a serious challenge around the world, with almost 2/3 of the global population experiencing water scarcity at some point during the year, especially in dry regions. One solution to this problem is sorbent-based atmospheric water harvesting (SAWH) due to its ability to produce drinking water in a range of environments, including at low humidity. SAWH device operation is composed of adsorption and desorption phases. During adsorption, moist air flows into the device and is adsorbed onto the sorbent bed. This is followed by the desorption phase during which the sorbent is heated to desorb the water as vapor, which is then transported to a colder condenser surface on which it is condensed as liquid water. Finally, the condensed water can be collected outside the device. However, current state-of-the-art SAWH devices are inefficient, with less than 70% of their adsorbed water being collected. This means the adsorbed water is either not condensed or condensed but not collected. This work discusses the impact of the coupling between desorption and condensation on the efficiency of SAWH devices. In general, SAWH systems can suffer from three scenarios of inefficient desorption-condensation: flux-limited, when the desorption rate in the device is insufficient to fully utilize the condenser’s condensation capacity; transport-limited, when the time scale of the vapor transport from the sorbent bed to the condenser is slow compared to the desorption operation time; and condenser-limited, when the condenser has a poor thermal design compared to the vapor flux. We developed a system-level model of a SAWH device to inform design strategies to mitigate these three bottlenecks and optimize device performance. Additionally, we quantified hydrocarbons, common airborne contaminants, as a mechanism for slowing water collection. Experimental findings are used to develop a model for the impact of airborne hydrocarbon adsorption on surface wettability and water retention for six metals commonly used as condenser materials. The findings from these models can inform design recommendations for SAWH devices as well as various other industrial applications in which water condenses on metal surfaces such as refrigeration and power generation. Future work will focus on continued experimental validation of the models.

Impact of Vimentin Intermediate Filaments on 3D Multicellular Collective Behavior

Thu, 01 May 2025 00:00:00 GMT

Impact of Vimentin Intermediate Filaments on 3D Multicellular Collective Behavior Rodriguez, Camille Dyani Vimentin, a type III intermediate filament, is an understudied component of the cytoskeletal system. However, in recent studies we can see its structural and mechanical properties aid in a cell's survival and migration. It forms a hyperelastic network and works synergistically with actin and microtubule to protect against large deformations. Despite vimentin intermediate filaments critical role in many biological processes, there are limited studies on its role in collective migration in 3D in vitro. To elucidate vimentin’s role in a collective cell cluster, single MCF-7 cells are embedded in a Matrigel-Alginate gel, which then grow into multicellular systems. The MCF-7 cells utilized are vimentin null, chemically inducible to form vimentin networks that interact with the other components of the cytoskeleton. These MCF-7 allow for controlled expression of mature vimentin intermediate filament (VIFs) which then form networks. We study these multicellular clusters over the course of 14 days. We demonstrate that there are key differences in morphology and mechanics, with the presence of vimentin. Our results suggests VIFs create more irregular cell clusters with more visible dynamic interplay with the environment. Uninduced (no VIFs) clusters were overall less dynamic and exhibited spherical morphology and minimal protrusions. Cluster with mature VIFs tended to form more elongated multicellular clusters with increased number of projections into the surrounding gel. In these induced (with VIFs) clusters these projections are shown to be constantly protruding and retracting along with the nuclei continually reorganizing. Our results show that these projections are accompanied with increased protrusive and contractile gel displacements. These results indicate that vimentin network generate an dynamic and functional morphology, along with mechanically perturbing their environment in the early stages of cell cluster collective behavior.

Planning for Dynamic Nonprehensile Object Transport

Thu, 01 May 2025 00:00:00 GMT

Planning for Dynamic Nonprehensile Object Transport Wang, Eric K. Generalized planning methods for dynamic manipulation struggle to efficiently solve kinodynamic constraints. Gradient-based methods suffer from initialization sensitivity, local optimum convergence, and lack of feasibility guarantees, while sampling-based methods can require large computation times if there exist challenging boundary conditions. Iterative Time Optimal Path Parameterization, or iTOPP, guarantees a feasible local minimum for a dynamic grasping problem by iteratively decreasing transit time for a trajectory initially generated to satisfy kinodynamic contact constraints. We demonstrate solutions that can handle initial or final goal states defined as quasistatically infeasible, in which purely quasistatic motions cannot generate a warm start trajectory. We also design an indirect adaptive controller that can track a desired dynamic grasping trajectory assuming unknown object mass and location parameters.

Fundamental Behavior of Nanoporous Networks in the Out-of-Autoclave Manufacturing of Carbon Fiber Reinforced Polymer Composites

Sun, 01 Sep 2024 00:00:00 GMT

Fundamental Behavior of Nanoporous Networks in the Out-of-Autoclave Manufacturing of Carbon Fiber Reinforced Polymer Composites Webb, Alisa Nicole Throughout the aerospace industry, carbon fiber reinforced polymer (CFRP) laminated composites are used extensively in spacecraft and aircraft vehicles due to their high specific strength and stiffness and other properties. Processing these advanced structural CFRP composites, especially in prepreg form, is often completed via autoclaves where elevated temperatures and pressures of typically 180 ◦C (350 ◦F) and 0.7 MPa (7 bars), respectively, are applied to cure the polymer matrix and compress the constituent laminae together. However, autoclaves are energy intensive, expensive, and impose geometrical constraints on components due to thermal gradients within the chamber. Thus, there exists a need to find alternative manufacturing techniques. Throughout this thesis, an alternative method to autoclave processing is presented using vacuum-bag only (VBO) techniques with nanoporous networks (NPNs) in the interlaminar regions in autoclave-required epoxy prepreg CFRP composites. Nanoporous materials are defined as materials containing pores in the mid nanometer to low micrometer range. Once placed in the interlaminar region of the laminate, voids are reduced by the induced capillary pressures of the NPNs, and trapped gas evacuates through the NPN. By utilizing capillary flow porometry, capillary pressure and through-thickness permeability are quantified for various NPNs, along with other porous materials. Capillary pressure and permeability exhibit an inversely proportional relationship for all tested materials with CNT-based and polymer aerogel NPNs providing capillary pressures higher than an autoclave pressure of 0.7 MPa. Accordingly, an Ashby-type plot is presented as an aid for NPN selection for composites manufacturing. Previous studies of unidirectional glass fiber reinforced polymer (GFRP) composites and unidirectional CFRP composites show success with NPN-enabled VBO-manufacturing using aligned carbon nanotubes (A-CNTs) and electrospun polymer nanofiber (EPN) mats. However, success with woven prepreg has not been consistently achieved before this thesis. Autoclave woven epoxy CFRP laminates of IM7/8552 are manufactured using EPN and polymer aerogel NPNs with a VBO procedure. Once manufactured, these laminates were characterized for quality through void content analysis. 0.11 void vol% was achieved which is well within the 1 vol% of void requirement for aerospace-grade composite components. To aid the in the understanding of NPNs, in situ experiments utilizing microcomputed tomography are developed to investigate the (presumed Newtonian) flow of resin throughout the NPN as a function of temperature, which varies throughout a typical manufacturer recommended cure cycle (MRCC), along with the void evolution throughout the cure cycle. Based on this new in situ understanding, a manufacturing process modification is devised to produce void-free woven laminates at the 152.4 mm laminate scale. Through manufacturing, material characterization, and designed in situ experiments, this thesis demonstrates the use of NPNs for VBO-manufacturing of low-void content aerospace-grade CFRP composites to replace autoclaves for energy and cost savings.

Mediators: Participatory Collective Intelligence for Multi-Stakeholder Urban Decision-Making

Thu, 01 May 2025 00:00:00 GMT

Mediators: Participatory Collective Intelligence for Multi-Stakeholder Urban Decision-Making Gao, Jin Cities are dynamic and evolving organisms shaped through the check-and-balance of interest exchange. As cities gain complexity and more stakeholders become involved in decision-making, reaching consensus becomes the core challenge and the essence of the urbanism process. This thesis introduces a computational framework for AI-augmented collective decision-making in urban settings. Based on real-world case studies, the core decision-making process is abstracted as a multiplayer board game modeling the check-and-balance dynamics among stakeholders with differing values. Players are encouraged to balance short-term interests and long-term resilience, and evaluate the risks and benefits of collaboration. The system is implemented as a physical interactive play-table with digital interfaces, enabling two use cases: simulating potential outcomes via AI self-play, and human–agent co-play via human-inthe-loop interactions. Technically, the framework integrates multi-agent reinforcement learning (MARL) for agent strategy training, multi-agent large language model (LLM) discussions to enable natural language negotiation, and retrieval-augmented generation (RAG) to ground decisions in contextual knowledge. Together, these components form a full-stack pipeline for simulating collective decision-making enriched by human participation. This research offers a novel participatory tool for planners, policymakers, architects, and the public to examine how differing values shape development trajectories. It also demonstrates an integrated approach to collective intelligence, combining numerical optimization, language-based reasoning, and human participation, to explore how AI–AI and AI–human collaboration can emerge within complex multi-stakeholder environments.

Creating space for HVAC systems: A new, intuition-building approach to HVAC system integration in architectural education and practice

Thu, 01 May 2025 00:00:00 GMT

Creating space for HVAC systems: A new, intuition-building approach to HVAC system integration in architectural education and practice Irani, Ali Heating, Ventilation, and Air Conditioning (HVAC) systems are vital to ensuring a healthy indoor environment in buildings. They are essential to the global shift toward a decarbonized, all-electric future. While integrated design practice has promised cost, energy, and space savings due to earlier and more frequent collaboration between design disciplines, remaining missed opportunities in the HVAC system design and coordination process often lead to spatial conflicts, performance tradeoffs, and uncomfortable spaces. This dissertation aims to understand current coordination practices to identify the root causes of existing problems, timeline issues, and knowledge gaps. Then, it proposes a series of enhancements to address these shortcomings, focusing on National Architectural Accrediting Board (NAAB) accredited architectural education programs that train the next generation of practicing architects. The proposed research hypotheses are validated in a three-part research approach: (1) releasing architecture industry surveys and conducting interviews, (2) designing and testing an early-stage design tool, and (3) developing, implementing, and evaluating a comprehensive HVAC curriculum for architecture students. The dissertation demonstrates that with the right tools and educational resources, architecture students can make informed, intuition-based HVAC system selections and integrate them into their building design, with students who studied the comprehensive curriculum demonstrating a 13% improvement in understanding and application of HVAC concepts compared to a control group of students. This work helps bridge the knowledge gap regarding HVAC systems, empowering designers to coordinate more effectively and prioritizing the role of HVAC systems in building performance simulation education.

From Scar To Scaffold: The Afterlife of the Oil Pipeline for a Decarbonizing World

Thu, 01 May 2025 00:00:00 GMT

From Scar To Scaffold: The Afterlife of the Oil Pipeline for a Decarbonizing World Apostolopoulou, Katerina With over 86,000 kilometers of crude oil pipelines—and more than 2.13 million kilometers of total oil and gas pipelines in the United States as of 2024—many segments are already corroded and aging, deeply embedded within urban and ecological systems that are increasingly endangered. As the global energy transition accelerates, this thesis investigates the future of these infrastructures, reconsidering the vast network of decommissioned and declining legacy pipelines not as obsolete relics, but as latent spatial assets for ecological repair, climate resilience, and socio-environmental justice. Moving beyond narratives of extraction and decay, the project repositions pipelines as linear territories of opportunity—capable of being retrofitted into new civic, ecological, and infrastructural frameworks. Central to the project is the transformation of the pipeline’s linear, extractive logic into a circular and connective one: a loop that is both finite and infinite, territorial and experiential. Focusing on a strategically selected loop of crude oil pipelines spanning 14 states, the thesis constructs a cartographic and architectural framework to reimagine these lines as sites of ecological repair, social infrastructure, and alternative energy distribution—where design, much like a biological scaffold, acts as a catalyst for regeneration along landscapes shaped by extraction. Through spatial analysis, typological classification, and mapping, five territorial conditions are defined along the pipeline loop, each offering distinct opportunities for intervention. These are tested through speculative design prototypes that transform the pipeline through operations of repurpose, renewable energy distribution, or ecological remediation. The interventions reframe invasive infrastructures into public and environmental assets—generating new spaces for inhabitation, production, and collective memory. Ultimately, the thesis proposes a post-carbon design paradigm rooted in ecological reciprocity, collective agency, and infrastructural care—revealing hidden energy landscapes and inscribing them with new values: resilience, equity, and repair.

Control and Aerodynamic Design of a Solar Road Vehicle with Articulated Surfaces

Thu, 01 May 2025 00:00:00 GMT

Control and Aerodynamic Design of a Solar Road Vehicle with Articulated Surfaces Salmon, Jason The automobile industry is critical to modern society. Simultaneously, the constant release of toxic emissions such as greenhouse gases into the atmosphere is detrimental to health and the environment. Vehicles which exploit cleaner energy sources would be preferable to reduce the horrific scale of human-initiated damage such as climate change. However, solar road vehicles—though designed and fabricated by some—have not reached a sufficient level to be production-worthy. The low efficiency of solar cells and the high energy demands of the average land vehicle are irreconcilable for most manufacturers using industry methods and design precedent. Therefore, this work centres around the design and control of a solar road vehicle which fundamentally breaks from the mould of the typical road vehicle design—a vehicle which employs extensive articulated surfaces (dubbed "solar wings") which can be angled to directly face the sun, thereby maximising solar irradiation. A solar tracker using Bayesian inference achieving promising results in both convergence and accuracy is presented. Additionally, a systematic method for optimizing a solar road vehicle with solar wings is developed and documented.

Mechanisms of Multi-Object Working Memory and Motion Prediction in the Primate Brain

Thu, 01 May 2025 00:00:00 GMT

Mechanisms of Multi-Object Working Memory and Motion Prediction in the Primate Brain Watters, Nick Sample-efficient learning and flexible generalization are hallmarks of intelligent behavior. Both sample-efficient learning and flexible generalization rely on re-using a mental model of the world in new contexts. For many decades, researchers in cognitive science, neuroscience, and machine learning have studied competing theories about the structure of our mental model of the world. One set of theories concerns the structure of multi-object representations in the brain. Some studies claim the brain represents multiple objects by allocating them to disjoint “slots” in working memory, others claim that the brain flexibly distributes a common pool of resources across objects, and yet others claim the brain represents multiple objects by rapidly switching between them through time. Another set of theories concerns the nature of predicting object motion. Some claim that the mind has an internal model of physics in the world that it uses to simulate the motion of objects through time, whereas others claim the mind relies on priors and heuristics to predict object motion without explicit simulation. Both of these sets of competing theories are long-standing and unresolved. In this work, we tackle these two open questions using primate neurophysiology and computational modeling. We trained monkeys to perform multi-object memory and motion prediction tasks, recorded large-scale single-unit activity from frontal cortex brain areas, and rigorously compared different hypotheses for the neural mechanisms of multi-object working memory and motion prediction. In the case of multi-object working memory, we found that the neural activity we recorded is more consistent with a model that flexibly distributes attentional resources across objects than with models that use object slots or temporal switching representations. In the case of motion prediction, we found that the neural activity is not consistent with the monkeys simulating an occluded moving object in real-time. Instead, the monkeys’ neural activity is driven largely by an anticipation of the position of the object at a future point in time. Both of these findings call into question long-standing cognitive theories and imply that the brain’s model of the world incorporates attentional mechanisms, priors, and heuristics. Lastly, we introduce a neural data preprocessing method for stabilizing electrophysiology recordings. This method improves spike-sorting results, helped us recover more neurons from our data, and we hope may help others make the most of their electrophysiology data as well.

Design and Commercialization Strategy of a Gantry-Based Automation Platform for High-Throughput Raman Spectroscopy

Thu, 01 May 2025 00:00:00 GMT

Design and Commercialization Strategy of a Gantry-Based Automation Platform for High-Throughput Raman Spectroscopy Romero, Catalina Raman spectroscopy is a powerful optical technique that enables rapid, label-free molecular analysis. This offers significant potential to be used across pharmaceutical development, microbiome research, and food diagnostics. However, the utility of Raman spectroscopy in high-throughput applications has been limited by the lack of cost-effective, modular automation platforms capable of handling large volumes of samples with precision and repeatability. Conventional Raman workflows are constrained by manual sample handling, slow throughput, and high user variability, limiting their applicability in high-volume testing environments. To address these challenges, this thesis presents the development and initial validation of a custom two-axis (XY) gantry and a robotic well plate stacker automation platform designed to streamline the sample handling workflow in Raman spectroscopy systems, facilitating high-throughput, precise, and reproducible positioning of microplate samples under a Raman microscope. This thesis also provides a commercialization framework for the system as a standalone automation product, targeting pharmaceutical high-throughput screening, microbiome analysis, and food safety testing. The platform serves the unmet needs in these industries, where labor-intensive and inconsistent sample positioning limits scalability. The commercialization analysis includes an evaluation of market sizing, competitive benchmarking, pricing models, and go-to-market strategies. The modular platform has the potential to enable broader adoption of Raman-based analysis tools by reducing labor intensity and improving repeatability in sample positioning workflows. This work lays the foundation for the future integration of optical feedback and automated analysis, with the goal of transforming how Raman-based diagnostics are conducted at scale.

Dishing It Out: Reimagining Multicultural College Dining Through Student-Centered Design

Thu, 01 May 2025 00:00:00 GMT

Dishing It Out: Reimagining Multicultural College Dining Through Student-Centered Design Dong, Annie Dining halls are central spaces in colleges, fostering not only nourishment but also cultural connection and community. However, when dining centers fall short in catering to the needs of their multicultural student body, students are often left feeling isolated and even further from home. Using MIT as a case study, this thesis employs user research and digital storytelling to explore how collecting student perspectives can inform college dining centers on better supporting the diverse cultural backgrounds and dietary needs of their students. The research and findings highlight the critical gaps and strengths in cultural representation within MIT’s dining halls. Through surveys and user research, this thesis gathers student perspectives on food authenticity, comfort, and identity, which inform the design of an interactive website prototype exploring student culinary backgrounds and preferences. This project serves as both a resource for dining services and a digital cookbook curated by the student body. By centering student voices through a culinary lens, this project aims to reimagine dining spaces as inclusive, representative, and comforting shared spaces within college campuses.

Intelligence Through Interaction: Leveraging Physical Interactions In Simple Robots To Produce Complex Behaviors

Thu, 01 May 2025 00:00:00 GMT

Intelligence Through Interaction: Leveraging Physical Interactions In Simple Robots To Produce Complex Behaviors Spino III, Pascal This thesis investigates how intelligent robot behavior can emerge from physical interactions rather than sensing, computation, and actuation in the traditional sense. Two robotic systems are presented to explore this concept in different domains. The first is a swarm of simple rolling robots whose collective morphology is shaped by distributed control laws and magnetic interactions, enabling decentralized construction-like behaviors such as bridge formation. The second is a soft underwater robot inspired by anguilliform swimming, which achieves efficient locomotion through a single actuator that leverages fluid–structure interactions in a compliant silicone tail. Useful behavior arises in both systems from the physical design and the dynamics of environmental interaction, rather than from algorithmic or computational complexity. These results demonstrate that physical intelligence can serve as a powerful design principle for building adaptive, robust, and minimal robotic systems.

Application of Revenue Management to Satellite Communications

Sun, 01 Sep 2024 00:00:00 GMT

Application of Revenue Management to Satellite Communications Eiskowitz, Skylar As the demand for satellite Internet continues to grow, satellite communication (SatCom) operators are faced with the challenge of effectively managing their capacity sales. While Revenue Management (RM) techniques have been widely used in other industries such as airline, hotel, and car rental services, the application of these methods in the context of SatCom is still scarce. This Thesis aims to bridge this gap by developing RM concepts, techniques, and optimization algorithms specifically tailored to the unique operational characteristics of SatCom capacity management and sales. The proposed SatCom RM method guides operators with quantitative recommendations of the amount of capacity to sell to different products in time and in different regions to maximize revenues. Though SatCom has characteristics that favor the use of RM concepts (perishable inventory, fixed capacity with a low variable cost, the possibility to segment demand), there are unique structural characteristics that complicate the development of SatCom RM models. The primary challenge is that different products consume varying amounts of capacity, with larger terminal size products utilizing less power on a satellite than smaller terminal size products. Moreover, the selling practices in SatCom are complex because products may be sold in one period and consumed across multiple periods in which additional sales are made. This requires rolling both the selling and consumption periods. Lastly, the SatCom RM problem poses a multidimensional network problem, as products can consume bundles of resources in both space and time. We extend two commonly used airline RM algorithms, Expected Marginal Seat Revenue (EMSRb) heuristic and Displacement Adjusted Virtual Nesting (DAVN) to the SatCom problem to create booking limits. The booking limits recommend a threshold amount of capacity an operator should sell of each product. The contribution of this Thesis is the modification of established airline RM algorithms to handle products with variable capacity uptakes. Further, these algorithms typically account for displacement costs of products, but only in one dimension of space or time (e.g., selling an airline flight that uses multiple spatial legs may displace capacity away from flights that only use one leg). Our modifications allow for the consideration of displacement costs in both dimensions of space and time. In order to evaluate the effectiveness of our inventory control approach, we conduct simulations of various demand scenarios and compare the revenue gains to a baseline scenario with no controls, as well as a simpler method that does not consider product duration. In a large-scale simulation spanning three years and encompassing thousands of product requests, we observe revenue gains ranging from 15%-30% depending on the demand scenario. Then, we extend the model to multiple zones and achieve 2%-10% revenue improvement using our Multi-Zone DAVN method compared to the DAVN method applied to each zone separately.

Hidden Monuments

Thu, 01 May 2025 00:00:00 GMT

Hidden Monuments Lee, Sesil Jeju Island’s burial culture is embedded in the island’s distinct landscape, where sandam burial mounds are not isolated monuments but quietly coexist with fields, ranches, and forests. These sites are living records of intangible heritage—ancestral beliefs, Beolcho rituals, and vernacular stone-stacking practices—manifested not through formalized memory, but through their modest yet persistent presence in the landscape. Today, however, these spaces are under threat: policies favoring cremation, rapid urbanization, and shifting land values render them increasingly invisible or obsolete. In the past few decades, two-thirds of sandam have been displaced, and with fewer than six out of over 100,000 burial sites designated as cultural heritage, traditional models of conservation are inadequate—unable to engage with the dispersed, landscape-bound nature of these burial grounds. This project reimagines Jeju’s burial mounds not as relics to be preserved, but as spatial anchors for cultural and communal expressions. Through a series of small-scale architectural interventions—gates, stages, passages, and shelters—deployed along paths tracing sandam clusters, the work explores how memory can be practiced rather than displayed. By offering ways to engage with the buried, the forgotten, and the living simultaneously, the project expands the idea of heritage: not as a static record, but as a participatory and evolving relationship between people, land, and memory.

Non-invasive tuning of experience-dependent plasticity in the primary visual cortex

Thu, 01 May 2025 00:00:00 GMT

Non-invasive tuning of experience-dependent plasticity in the primary visual cortex Reilly-Andújar, Francis The cerebral cortex exhibits a remarkable capacity for experience-dependent plasticity, a feature that is predominantly confined to critical periods (CPs) during early postnatal development. In the mouse primary visual cortex (V1), ocular dominance plasticity (ODP) has served as a premier model for investigating the cellular and molecular mechanisms that underlie the formation and stabilization of cortical circuits. During the CP, short-term monocular deprivation (MD) induces both functional and anatomical changes in binocular V1, characterized by a weakening of deprived-eye responsiveness via mechanisms of synaptic long-term depression. As the critical period closes, increased inhibitory drive and the emergence of perineuronal nets (PNNs) stabilize neural circuits and restrict further experience-dependent plasticity. In Chapter 1, I review the key literature on ODP and provide a survey of interventions that have been shown to enhance ODP in adulthood. In Chapter 2, I present our findings that repeated anesthetic ketamine treatment can reinstate ‘juvenile-like’ plasticity in the adult mouse V1. Importantly, I demonstrate that this effect relies on the microglia-mediated depletion of PNNs, and that interfering with microglial purinergic P2Y12 receptor activation blocks the ketamine-induced enhancement of ODP. Building on these insights, Chapter 3 investigates the use of non-invasive light-flicker stimulation at different temporal frequencies as a means to unlock different forms of ODP in the adult mouse V1. Our results reveal that 60 Hz light-flicker stimulation reduces PNN levels and promotes a depression of deprived-eye responses following short-term MD, whereas 40 Hz stimulation – without altering PNN levels – enhances an adult form of ODP characterized by the strengthening of non-deprived eye responses following short-term MD. Furthermore, we show that in mice subjected to long-term MD initiated early in life, 40 Hz light-flicker treatment promotes recovery of visual function, as evidenced through physiological and behavioral assays. Finally, Chapter 4, outlines a series of future experiments designed to further elucidate the mechanisms by which light-flicker stimulation promotes enhanced ODP in adult V1. Together, the findings presented in this thesis introduce novel, minimally invasive (ketamine) and non-invasive (light-flicker) interventions that show promise as therapeutic strategies for ameliorating deficits arising from early life sensory deprivation.

Development of Mechanical and Electrical Interfaces for Rapid Swap Battery Systems

Thu, 01 May 2025 00:00:00 GMT

Development of Mechanical and Electrical Interfaces for Rapid Swap Battery Systems Wucherer, Abigail In the drive towards a globally decarbonized energy economy, rapid swap battery packs provide a potential means to improve electric vehicle adoption in high utilization industrial vehicles where lengthy charge times are a barrier to electrification. High voltage, high current battery connectors are a critical component for coupling the pack to the electric vehicle, distributing power from the battery to the drivetrain. Most state-of-the-art connections require precision alignment of contact surfaces, and bolted preload or retention mechanisms, hindering the implementation of rapid swap battery systems. The need for robust, high life cycle, high-power contacts motivates a new approach to connector design. The integration of electrical connectors with the battery mount’s structural loop creates a new design space where preload, geometry, and contact resistance may be optimized. This co-design approach enables mechanical and electrical functional requirements to be considered in conjunction to ensure reliable fulfillment in both areas while reducing the time for battery pack swaps. This work introduces two distinct approaches for aligning the pack to the vehicle, locking the battery in place, and engaging electrical contact with geometry unique to the system design. These approaches offer higher reliability, mechanical and electrical longevity, and automatic alignment capabilities during loading of the battery pack. Across both designs, the contact resistance is the primary metric for evaluating the electrical performance, and the contact pressure is used to evaluate the risk of mechanical wear. The first approach integrates a quasi-kinematic coupling-based connector with integrated electrical contacts, allowing for repeatable and accurate positioning of the battery pack to the vehicle. A slotted ball and socket design approach is considered to accommodate for angular misalignment and establish repeatable contact area through elastic averaging. The second approach proposes a planar contact to further reduce the contact pressure and increase contact longevity without the need for expensive and rare hardened coatings. This system relies on a rail and flat system for guiding the battery pack into a locked position and engaging the planar contacts.

Weaving Borders, Mapping Place: Afghan War Rugs of the Soviet-Afghan War (1979-1989)

Thu, 01 May 2025 00:00:00 GMT

Weaving Borders, Mapping Place: Afghan War Rugs of the Soviet-Afghan War (1979-1989) Hakemy, Arezo Early Afghan war rugs delineate place through their pictorial design, embedding spatial memory into the tactile surface of the woven field. Emerging in the wake of the Soviet invasion in the late 1970s, these rugs integrate modern war iconography of tanks, helicopters, and maps into a medium historically tied to regional identity, spiritual practice, and craft. While earlier scholarship has often read these rugs as commodities of war tourism, this thesis moves beyond this interpretation to foreground the rug as a placemaking device, one that asserts territory and agency through mapping techniques. Afghan war rugs frame and define space on a land that has largely been considered placeless, at times porous and seemingly unknown. Through their borders, these rugs resist the geopolitical narratives that have long reduced Afghanistan to a war-torn frontier. The border serves as a framing device, structuring the rug’s design while simultaneously asserting territorial presence. Whether following a prescribed cartoon or improvising patterns, the weaver actively engages in “border-ing,” exercising cartographic agency by embedding personal, traditional, and political motifs into the rug. This research interrogates how early Afghan war rugs engage in spatial representation against the backdrop of the Soviet-Afghan war from 1979-1989. From historical colonial mapping projects to Soviet and American cartographic investigations, Afghanistan’s borders have long been sites of surveillance, resource extraction, and imperial ambition. Yet, in contrast to these external mapping practices, the war rug’s design is a resistant act of placemaking. Examining the rug as both artifact and map, this study explores how Afghan weavers reclaim their landscapes through rug making, embedding memory and materiality into woven form.

The role of texture in auditory scene analysis

Thu, 01 May 2025 00:00:00 GMT

The role of texture in auditory scene analysis Hicks, Jarrod M. Everyday auditory scenes contain sounds from many sources. For example, when crossing the street, you might hear sounds produced from the rumble of passing cars, the chatter of pedestrians, and the rapid tick of crosswalk signals. To make sense of this complex mixture of sounds, the auditory system must separate the mixture into coherent perceptual representations that are likely to correspond to the underlying sources in the world. This process is known as auditory scene analysis. Although a rich body of work has probed auditory scene analysis with simple synthetic stimuli and revealed principles of perceptual organization, the extent to which these principles apply to real-world scenes with natural sounds remains unclear. This thesis empirically examines auditory scene analysis with realistic sounds. In particular, we study the perception of scenes containing a common class of environmental sounds known as “textures”, investigating how the auditory system makes use of statistical structure to separate textures from other sources and how the underlying statistical representation both constrains and enables scene analysis. We first investigated the mechanisms of hearing in noise using real-world background “noise” textures. The results show that the auditory system estimates the properties of “noise” textures and stores them over time, using the resulting internal model to estimate other concurrent sounds. We then considered how concurrent sound texture sources are separated from each other. We found that auditory scene analysis with textures involves some principles identified in classical scene analysis work with simple sounds, but that these principles apply to the higher-order statistical representations that define natural textures. Together, the results reveal new aspects of auditory scene analysis with real-world sounds and clarify the role texture plays in everyday hearing. Our findings provide a bridge between the simple, synthetic stimuli studied historically and the rich complexity of real-world sounds.

Driving Temporally Precise Learning in Individual Premotor Neurons using Closed-Loop Neurofeedback

Thu, 01 May 2025 00:00:00 GMT

Driving Temporally Precise Learning in Individual Premotor Neurons using Closed-Loop Neurofeedback Scherrer, Josefa R. Much of human existence is based on our ability to learn complex sequences of motor movements. Speech, writing, and tool use all require activating a series of different muscles in a precisely timed pattern, and these patterns are learned through a long process of trial and error. How does the neural circuitry in our motor system learn to generate the activity patterns that drive these sequences? This question can be explored by studying a similarly precise learned motor pattern in a different organism, the learned song of the songbird zebra finch. Zebra finches learn to sing a stereotyped song through a process of vocal experimentation and comparison to an internal template. Every time a bird sings, it varies the acoustic parameters of its song and determines whether each variation brings the song closer to its internal template. Variations that result in a better match are then repeated in subsequent renditions of the song, in a trial and error process suggestive of reinforcement learning. The learning process requires a basal ganglia-thalamocortical loop called the anterior forebrain pathway (AFP) that is similar to basal ganglia-thalamocortical circuitry in mammals. Existing evidence suggests that the AFP learns a time-dependent bias signal that steers the motor pathway to avoid vocal errors. This bias signal is known to be dependent on the cortical output of the AFP known as LMAN (lateral magnocellular nucleus of the anterior nidopallium). However, little is known about the neural code in LMAN that underlies this bias signal, or how this neural code is learned and generated. We address these questions by building a neural feedback system that allows us to impose correlations between the activity of individual LMAN neurons and a dopaminergic reward signal. We designed a low-latency feedback system that records neural activity from a chronic Neuropixels 2.0 implant, extracts the activity of specific neurons, and plays noise bursts to the bird contingent on the activity of those neurons. We used this system to perform feedback based on the activity of an arbitrarily chosen neuron in LMAN within a given 10 ms window in songs. All birds responded to the feedback by learning to bias the activity of the chosen LMAN neuron up or down within the chosen time window, transiently driving firing rates up by as much as 200 Hz. We observed a remarkable degree of timing precision in the learned bias, with birds able to control the activity of the chosen neuron at single millisecond levels of rise time and jitter. This high degree of precision informs models of the basal ganglia circuit architecture thought to drive learning. We also found the learned bias to be specific to the LMAN neurons correlated with reward, with neighboring uncorrelated neurons exhibiting no change in firing rate during learning. This single-neuron specificity strongly constrains the spatial precision of axonal targeting from thalamic regions that are thought to propagate the learned bias signal from the basal ganglia to LMAN. Finally, we demonstrated that fluctuations in neural activity of a given LMAN neuron drive transient and predictable changes in vocal output approximately 25 milliseconds later, consistent with what is known about signal propagation speeds in the song system. This fact, together with the results of our feedback experiments, combine to confirm our central hypothesis that LMAN drives song learning by independently activating LMAN neurons at precise points in time in order to bias vocal output and avoid vocal errors.

Social Sensory Somatic Scores for Species, Spaces, Soils, and Structures of Steep Slopes

Thu, 01 May 2025 00:00:00 GMT

Social Sensory Somatic Scores for Species, Spaces, Soils, and Structures of Steep Slopes Bondarenko, Lina Modern knowledge systems have physically and conceptually “flattened” the world, erasing the ecological, political, and sensory complexities inherent to sloped terrain. By attending closely to the slope—as both a material condition and a generative metaphor—this thesis foregrounds movement as a form of resistance to regimes of exploitation, abstraction, and estrangement that have historically transformed land into data and place into property. Weaving together interdisciplinary methodologies from performance studies, landscape architecture theory, feminist geography, ecological theology, environmental history, sensory ethnography, and media studies, SSSSSSSSSS dances an inclined methodological structure, oscillating deliberately between critical systemic analysis and situated sensory experience. Ch1. sets the stage among steep slopes and introduces the discipline to movement as pedagogy, enacting the urgency for new methodologies into schemes of the project’s medium and the book’s format. Ch.2 is a feminist investigation of the ways modern infrastructures and spaces have been designed to reinforce land abstraction and commodification in the name of improvement-- severing embodied relationality, contributing to societal apathy toward ecological and social crises. Imperial post-enlightenment statecraft, the suppression of wildness, and the standardization of level form have flattened our upright movements to enact a state of senslessness. Contradicting Ch.2’s straight critique, Ch.3 attempts to reweave the sinuous nuance of symbiogenesis between soils and species, revealing that humans are but one among many sloped organisms moving, and inclining, and co-evolving as the lithosphere; we have been slorgs all along. Slorgs belong to divine mythologies of terrain’s elevations and have reciprocated in admiration, mimicking topographic spatial functions and adorning the summits with artistic interventions--some inadvertently contributing to the damaging regimes of Ch.2. Interwoven through both chapters, outliers resisting those forces of governance and exploitation are often those displaced by them-- those moving in ways the system polices and erases from comprehension-- refugees, queers, witches, tricksters, artists, herbalists, and healers. The intended medium of SSSSSSSSSS coalesces in Ch.4: inviting the general public to participatory happenings with hills, composing scores, coaxing their inner slorgs to slither askew, sloping themselves as moving loci for sympoietic becoming. Multi-species attune to a social, sensed, somatic experience, co-composing spatial relations among local steep soils. Slorgs challenge the abstractions of dominant epistemologies in the temporal, situated act of trusting their own proprioception in collective balance, affirming the multidimensional value of embodied, ecological geo-choreography. Social Sensory Somatic Scores for Soils, Structures, Spaces, and Species of Steep Slopes are presented through photographs in Ch.4 and in moving image, available as supplemental material.

The Spiritual Curation of American Modernism

Thu, 01 May 2025 00:00:00 GMT

The Spiritual Curation of American Modernism Saha, Indrani Where do the spiritual go? In this study of late-nineteenth- and early-twentieth-century seekers, they join seances in Vermont farmhouses, attend Theosophical lectures on Karma, get lost in copies of Jnana-Y oga, journey to Buddhist temples in China, and consume spiritual manuals on Mentalphysics. But where do they go after those encounters? And, more importantly, what do they do? In this dissertation, they build modern art institutions. A cadre of artist-writers, museum curators, and public intellectuals found their power in early 20th century America by building institutions to introduce a new, spiritually grounded modern art to a mercantile nation. In the US, beyond European sources for "the spiritual" were flirtations with vaguely "Eastern" ones by way of Theosophy. Those who sought to institutionally manifest Wassily Kandinsky's "spiritual" in art believed themselves to provide the assistance necessary to cultivate and preserve these spiritual impulses in modern art. Alfred Stieglitz's Intimate Gallery (1925-1929), Katherine Sophie Dreier's Societe Anonyme (1920-1950), and Hilla Rebay's Museum of Non-Objective Painting (1939-1952)-all in New York City-served as intermediaries in translating predominantly Eastern spiritual ideas into productive ways of being. It would be needed, each curator believed, to cultivate these spiritual protocols just to survive in a material world they held to be detrimentally bankrupt of spirit. In other words, the American institutionalization of modernism built its canon around spiritual systems of national aesthetic welfare. Crucial to these spiritual curators' respective operations would be the promotion of not just any abstraction but a radically non-objective art thought to use the inner expressions of the artist to elevate the spectator. This dissertation takes the turn-of-the-century claims of spirituality by the founders of key art institutions seriously. In doing so, I argue that esoteric forms of Eastern spirituality infused formerly Protestant centers of culture to propel a twentieth-century embrace of radically abstract modern art.

Programmable Mud: 3D Printing earth to achieve low-carbon, low-cost construction automation

Thu, 01 May 2025 00:00:00 GMT

Programmable Mud: 3D Printing earth to achieve low-carbon, low-cost construction automation Curth, Alexander (Sandy) McCormick Large-scale additive manufacturing (LSAM) with locally sourced materials, such as earth, presents a promising approach to addressing the urgent challenges of rapid urbanization and construction-related carbon emissions. This dissertation establishes a comprehensive framework for integrating low-carbon materials, particularly minimally processed earth, with computational design methodologies and robotic fabrication processes for architectural-scale applications. Through systematic material characterization, novel testing protocols, and case studies across multiple building systems, the research demonstrates that minimally processed earthen materials can be transformed into high-performance building elements uniquely suited to local environmental conditions and design considerations. The developed computational framework employs multi-objective optimization and material-aware toolpath generation to balance structural performance, thermal comfort, embodied carbon, and construction time. Four case studies validate this approach: (1) toolpath optimization for shell structures, (2) a hybrid floor system combining shape-optimized concrete beams with 3D-printed ceramic blocks, (3) zero-waste earthen formwork for reinforced concrete, and (4) thermally optimized wall systems for passive climate control. Life cycle assessment reveals that 3D-printed earth structures have approximately one-fifth the embodied carbon of conventional concrete and one-fiftieth that of industry-standard 3D-printed mortar. This research bridges the gap between additive computational design and material circularity, offering scalable approaches to sustainable construction that can be implemented across diverse environmental and economic contexts.

Cooling Machines: Exploring the Heat Mitigation Effect of Urban Trees with Computer Vision

Thu, 01 May 2025 00:00:00 GMT

Cooling Machines: Exploring the Heat Mitigation Effect of Urban Trees with Computer Vision Klimenko, Nikita As the impacts of climate change on cities become more pronounced, urban authorities are under pressure to prepare existing streetscapes for increased levels of heat stress. While many aspects of existing urban morphology have an impact on heat exposure (e.g. sky view factor, glazing levels, facade materials), they cannot be rapidly changed at large across existing urban infrastructures. Urban authorities across the world increasingly turn to planting trees as a way of cooling urban streetscapes. Urban vegetation is indeed known to have a cooling effect, primarily due to trees providing shade and preventing urban materials from heating up, as well as due to their ability to maintain their own internal temperature due to evapotranspiration. Even though the positive impacts of urban trees on thermal comfort are long known and well-studied, little work is dedicated to how these impacts vary across trees of different species and morphology. This is due to both the complexity of studying vegetation life cycles at sufficient scale, as well as due to the dispersed nature of the issue across disciplines of biology, urban climate, design, and data science. Nevertheless, this specific knowledge is vital to urban planners for deciding which trees have the most cooling effect in specific parts of the city. This thesis embraces the notion of trees as ‘cooling machines’ and dissects the diverse morphological and contextual factors that affect the role of individual trees on local urban heatscape. Leveraging a set of computer vision methodologies, including species recognition, context-aware segmentation, and photogrammetry, the thesis examines a large dataset of thermal imagery of urban trees collected in Los Angeles and Dubai to describe the impact of individual tree species, height and form, as well as spatial context on the cooling effect. Building on this approach, the thesis proposes a prototyping framework for architects to cure urban heatscapes via targeted curation of tree planting schemes, tying the visual and thermal aspects of urban greenery. This approach will allow cities to leverage the power of urban vegetation in the most efficient way, and tame urban heat in a scalable and globally affordable manner.

Co-Authoring Beyond the Human: Disordering Architectural Processes through Play and Multi-Agent Co-Existence

Thu, 01 May 2025 00:00:00 GMT

Co-Authoring Beyond the Human: Disordering Architectural Processes through Play and Multi-Agent Co-Existence Dundar Arifoglu, Nasibe Nur This thesis reconsiders architectural authorship and the extended processes through which the built environment is shaped, using a series of playful, participatory interventions to expose the human-centric assumptions embedded in spatial decision-making. Presented as a collection of games and booklets, the work invites participants to engage with a wide spectrum of architectural processes—from site understanding and planning to permitting, construction, and post-occupancy—through the perspectives of multiple agents entangled in shared environments. These agents include beings, materials, living organisms, legal frameworks, and other forces typically excluded from spatial authorship, challenging conventional boundaries and expanding the discourse around the entangled forces and relations that shape the spaces we inhabit. A series of playful explorations opens space for friction, misalignment, and shared authorship. Each booklet engages a distinct stage of the architectural process through participatory formats that make visible the biases, exclusions, and regulatory fictions often treated as neutral. By gamifying these systems, the work reveals how architectural decision-making tends to privilege hierarchy, human control, and speed—often at the expense of multispecies co-existence. This thesis positions play as a critical lens: a way to rehearse alternative futures, to listen differently, to embody other perspectives, and to surface the black-box logics embedded in architectural norms. It invites readers and players to participate in unbuilding these assumptions. And the games evolve—with each use, each misreading, each encounter, and each agent who joins the conversation.

The Objectiles Guide to Time Travel: Re-Envisioning Building Materials as Narrative-Collecting Object-Projectiles on a Trajectory Through Space-Time

Thu, 01 May 2025 00:00:00 GMT

The Objectiles Guide to Time Travel: Re-Envisioning Building Materials as Narrative-Collecting Object-Projectiles on a Trajectory Through Space-Time Chaussabel, Celia Quynh-Mai As the architectural discipline grapples with its role in resource depletion, carbon emissions, and waste generation, there is a growing urgency to stop sourcing new materials and to reuse materials from existing buildings instead. One challenge to integrating reused materials into current building practices is technical: inventorying, deconstructing, reconditioning, and designing with reused materials is slower and more labor-intensive than with new ones. But another challenge is cultural: the materials that make up architecture are currently perceived as unmoving and single-use, with little consideration for their trajectories from raw resource to landfill. This thesis is focused on developing an aesthetic sensibility and design methodology that helps us re-envision materials as objects on a trajectory instead: Objectiles, or object-projectiles. Objectiles are objects on an adventure across space-time to collect as many uses as possible. Rather than remaining associated with one primary use, Objectiles are impressionable, bearing ambiguous traces of all the uses they encounter as they re-circulate. Through the aesthetic qualities that hint at their many uses, Objectiles invite us to time travel - to imagine the potential past and future narratives that may precede or follow their present physical state. Embedding the aesthetics of Objectiles into architecture can lead to the development of a new collective consciousness of the materials that surround us. They can make us aware that all the objects around us have trajectories that extend beyond their present state, and lead to an alternative material culture of greater care in how we use, re-circulate, and dispose of all objects.

Problem-Independent Regrets on Expectation-Dependent Multi-Armed Bandits

Thu, 01 May 2025 00:00:00 GMT

Problem-Independent Regrets on Expectation-Dependent Multi-Armed Bandits Ai, Rui The independence axiom (IA) proposed by Von Neumann and Morgenstern [50] is the cornerstone of the expected utility theory. However, some empirical experiments show that the IA is often violated in the real world. We propose a new kind of multi-armed bandit problem where the expectation of outcomes may influence the agent’s utility which we call expectation-dependent multi-armed bandits and rationalize the choice of agents in Machina’s paradox lacking the IA. We design provably efficient algorithms with low minimax regrets and show their consistency of time horizon T with corresponding regret lower bounds, revealing statistical optimality. Furthermore, as we first consider bandits whose corresponding utility depends on both reality and expectation, it provides a bridge between machine learning and economic behavior theory, shedding light on how to interpret some counterintuitive economic scenarios, like bounded rationality explored by Zhang et al. [54].

Differentially Private Synthetic Data Generation for Relational Databases

Thu, 01 May 2025 00:00:00 GMT

Differentially Private Synthetic Data Generation for Relational Databases Alimohammadi, Kaveh Existing differentially private (DP) synthetic data generation mechanisms typically assume a single-source table. In practice, data is often distributed across multiple tables with relationships across tables. This study presents the first-of-its-kind algorithm that can be combined with \emph{any} existing DP mechanisms to generate synthetic relational databases. The algorithm iteratively refines the relationship between individual synthetic tables to minimize their approximation errors in terms of low-order marginal distributions while maintaining referential integrity; consequently eliminates the need to flatten a relational database into a master table (saving space), operates efficiently (saving time), and scales effectively to high-dimensional data. We provide both DP and theoretical utility guarantees for our algorithm. Through numerical experiments on real-world datasets, we demonstrate the effectiveness of our method in preserving fidelity to the original data.

VLEO-Bench: A Framework to Evaluate Vision-Language Models for Earth Observation Applications

Thu, 01 May 2025 00:00:00 GMT

VLEO-Bench: A Framework to Evaluate Vision-Language Models for Earth Observation Applications Zhang, Chenhui Large Vision-Language Models (VLMs) have demonstrated impressive performance on complex tasks involving visual input with natural language instructions. However, it remains unclear to what extent capabilities on natural images transfer to Earth observation (EO) data, which are predominantly satellite and aerial images less common in VLM training data. In this work, we propose VLEO-Bench, a comprehensive evaluation framework to quantify the progress of VLMs toward being useful tools for EO data by assessing their abilities on scene understanding, localization and counting, and change detection tasks. Motivated by real-world applications, our framework includes scenarios like urban monitoring, disaster relief, land use, and conservation. We discover that, although state-of-the-art VLMs like GPT-4V possess extensive world knowledge that leads to strong performance on open-ended tasks like location understanding and image captioning, their poor spatial reasoning limits usefulness on object localization and counting tasks.

She Swims in Silence: Spatial Narrative, Women's labor in Contemporary Art

Thu, 01 May 2025 00:00:00 GMT

She Swims in Silence: Spatial Narrative, Women's labor in Contemporary Art Feng, Haozhen This thesis investigates the collective lives of Chinese women sent to Xinjiang in state-led migration after 1949 and the erasure of their gendered narratives. Drawing on a unique family history and archival evidence, the thesis reveals how the personal identities of these female “Aid to Xinjiang” participants were stripped away and subsumed under the grand socialist nation-building myth. Through practice-based artistic research, the project attempts to restore their lost voices and unacknowledged suffering and labor, framing the exhibition as a form of praxis. By analyzing the exhibition alongside case studies and critical analysis, the thesis, inspired by Bernard Stiegler’s theory of the “history of representational forms” and interwoven with ideas from philosophers like Judith Butler and Nicholas Mirzoeff, interrogates the gendered silences in official history and highlights the tension between state mythologies and personal memories. In doing so, the exhibition as an interdisciplinary form of research not only restores agency to a silenced group of women, but also demonstrates how artistic practice can serve as an alternative historiography to challenge dominant narratives and recover marginalized voices.

Evaluation of Universal Docking Solutions for Autonomous Underwater Vehicles

Thu, 01 May 2025 00:00:00 GMT

Evaluation of Universal Docking Solutions for Autonomous Underwater Vehicles Pryal, Erik Jeffrey Due to their energy-constrained nature, Autonomous Underwater Vehicles (AUVs) need effective docking and charging stations to extend their mission durations. However, diverse AUV designs challenge the universal compatibility of docking stations. This study provides a framework for understanding what makes a docking station universal and offers two potential solutions: the Tapered Funnel Docking Station and the Magnetic Hub Docking Station. The Tapered Funnel features a conical entry that progressively narrows to accommodate various AUV diameters. The Magnetic Hub passively secures the AUV using magnetic forces and an external appendage guided into position by a square duct. MATLAB simulations evaluate these two charging station designs for compatibility with AUVs, alignment capabilities, and docking efficacy under realistic conditions. Both designs are tested through Monte Carlo simulations to address varying AUV approach conditions, showcasing their potential as universally feasible solutions. Future exploration into material durability, sensor integration, and power transfer efficiency will refine these designs for field applicability. This research lays the groundwork for universal docking standards and proposes adaptable solutions to alleviate operational limitations in underwater missions.

Dowel-laminated timber from waste lumber offcuts: Towards structural component circularity

Thu, 01 May 2025 00:00:00 GMT

Dowel-laminated timber from waste lumber offcuts: Towards structural component circularity Blowes, Rachel In the context of the global climate crisis, there is a need to develop low embodied carbon building systems. Moreover, construction and demolition generate substantial amounts of waste. The use of salvaged materials for structural applications presents the opportunity to divert this waste while reducing the embodied carbon of new structural components. This thesis proposes a typology for dowel-laminated timber (DLT) slabs built up from waste lumber offcuts. A mechanical model for a segmented DLT system composed of geometrically heterogeneous offcuts is developed. Prototypes of this mass timber system are fabricated and tested to observe their failure behavior and to evaluate the mechanical model. A computational workflow is introduced which employs algorithmic methods for inventory assignment and structural optimization to design slabs which meet deflection requirements under loading. These approaches are undertaken to evaluate whether DLT systems can leverage the irregularity of salvaged lumber dimensions to produce structurally efficient forms.

Allopoietics in Real Time: Unfolding Among Art, Publics, Space, and Time

Thu, 01 May 2025 00:00:00 GMT

Allopoietics in Real Time: Unfolding Among Art, Publics, Space, and Time Aubry, Vinzenz This thesis proposes a conceptual lens for understanding contemporary generative arts by introducing the terms Allopoietics and Liquid Media. Building on generative and participatory art, it focuses on the real-time processes among artworks, publics, spaces, and time through which meaning dynamically emerges. Drawing on the author’s artistic works—Conjunktion, Looking at the Sun, and Public Eyes—as well as critical engagement with hermeneutics, process philosophy, and media theory, this thesis explores how agency is distributed across these processes, offering a means to reconsider all elements as equally generative. Allopoietics, derived from cybernetics, describes the generative capacity of systems to produce outcomes beyond the sum of their actants, emphasizing collective unfolding over isolated creation. Liquid Media expands the notion of interfacing beyond traditional media to include publics, space, and time, conceptualizing these as mutable and entangled actants. These concepts outline an Aesthetics of Real Time that evaluates the dynamic relations among increasingly immediate systems. By proposing these new terms, the thesis invites a shift in perspective from object to process: viewing artworks not as stable materializations but as parts of real-time systems of collective meaning-making. While emerging from an artistic practice, this conceptual framework resonates with insights from contemporary sociology and cultural studies, where notions of fluidity, distributed agency, and relationality increasingly shape our understanding of complex systems and realities.

The Limits of Longevity

Thu, 01 May 2025 00:00:00 GMT

The Limits of Longevity Rodriguez, Christopher W. Do all animals age? Although aging seems to be a widespread phenomenon, some demographic studies have failed to find evidence of aging in certain species, including some highly regenerative species of planarians and Hydra that reproduce through asexual fission. However, all demographic studies have limits on observation times and sample sizes, so it is unknown if these failures were because of an actual absence of aging or these inherent study limitations. Some argue that these species must be ageless. Because of pressures that result from the lack of a clean division between the germ line and the soma in fissiparous organisms, agelessness becomes necessary as a prerequisite of this kind of reproductive strategy. Others argue that fundamental theories of the evolutionary biology of aging absolutely preclude agelessness. Even putting evolutionary arguments aside, some mathematical models of cellular competition and senescence argue that agelessness is impossible mechanistically in multicellular organisms. In this work, I address evolutionary and mechanistic arguments for and against agelessness. I develop mathematical models of the Disposable Soma Theory that incorporate facets of the arguments for agelessness in asexual fissioning organisms. I construct models of mutation accumulation and drift within an individual and explore how this genetic decay could manifest in the mortality rates. I use these models to understand if aging is inevitable generally and apply them to planarians and Hydra to seek to estimate the likelihood of aging more narrowly in those specific cases. Contrary to other work, I find that agelessness (defined as non-increasing mortality rates in a population) is indeed possible as the optimal evolutionary strategy for multicellular organisms. However, the evolution and mechanistic realization of agelessness requires conditions that are unlikely to be met in any existing species. In the case of planarians and Hydra, they likely do not face the right kind of evolutionary pressure to completely avoid aging. Even if they do face necessary evolutionary pressure, intraindividual genetic decay will almost certainly induce increasing mortality on the population with little recourse. Therefore, these species likely do age, although they could have median lifespans on the order of hundreds or perhaps even thousands of years, which would make detecting aging in any given population study quite difficult indeed.

Fundamental representations of regions and interactions in spatial transcriptomics

Thu, 01 May 2025 00:00:00 GMT

Fundamental representations of regions and interactions in spatial transcriptomics Maher, Kamal M. While cells are often considered the fundamental unit of biology, it is their spatial coordination that gives rise to the tissue architectures underlying both health and disease. Spatial transcriptomics technologies offer a unique window into this coordination by simultaneously capturing the spatial and molecular identities of individual cells, providing unprecedented insight into tissue organization. However, the computational landscape for analyzing tissue structure remains fragmented, with a wide array of disparate methods. In this work, we aim to distill these approaches into a unified quantitative framework for analyzing tissue architecture. Tissue structure can be represented in terms of anatomical regions as well as the cellcell interactions that occur within them. For regional tissue organization, many existing methods—including those based on probabilistic models and graph neural networks—ultimately perform a form of smoothing, or local averaging of gene expression across neighboring cells. This process emphasizes large-scale spatial variation and enables standard single-cell analysis workflows, such as clustering and trajectory inference, to be applied in spatial contexts. However, we find that naive smoothing introduces artifacts that obscure meaningful spatial features. To address this, we introduce a minimal but powerful modification: subsampling within each neighborhood prior to averaging. This approach enhances spatial feature resolution and generalizes conventional analyses to spatial features: clustering identifies multicellular regions; data integration aligns spatial regions across samples and technologies; and trajectory inference captures spatial gradients. We also show that this subsampling strategy improves the performance of more complex downstream methods. To further generalize our framework, we formalize the joint analysis of tissue regions and multiscale cell-cell interactions using signal processing over graphs: low-frequency components represent regional gene expression patterns across a tissue mesh; high-frequency components capture fine-scale, cell-cell interactions; and mid-frequency signals correspond to boundaries between regions and diffusive signaling. By interpreting spatial gene expression in this spectral framework, we provide a principled way to bridge conceptual and computational perspectives on tissue structure. Ultimately, this work serves as both a theoretical foundation to understand existing methods and a roadmap for developing future approaches to quantitatively describe molecular tissue architecture.

Developing a Functional in Vitro Model of the Neuromuscular Interface

Thu, 01 May 2025 00:00:00 GMT

Developing a Functional in Vitro Model of the Neuromuscular Interface Schwendeman, Laura A. The neuromuscular system is responsible for the coordination of movement throughout the body, and while research has revealed many of the mechanisms involved in the function of the neuromuscular system, there are still many gaps in our understanding of how all of the components of the system work and how they are affected by environmental factors and disease. This work focuses on developing methods and an in vitro model for studying a subsystem of the neuromuscular system known as the neuromuscular junction (NMJ), which is the connection between skeletal muscle and motor neurons and is relevant in many neuromuscular degenerative diseases. This work identifies that current in vitro NMJ models are cohesively lacking the ability to support long-term, functionally contractile muscle tissue while providing compartmentalization and clear optical access for live imaging of muscle and motor neuron co-cultures. This work therefore presents STAMP, a microgroove patterning method for creating aligned, more physiologically relevant, functional, and optically accessible skeletal muscle tissue cultures on top of fibrin hydrogels. Through investigating a series of different sizing parameters, STAMP is shown to effectively align mouse and human skeletal muscle monolayers in vitro and influence the direction of muscle contraction under electrical and optogenetic stimulation while preserving skeletal muscle tissue integrity and viability. The STAMP approach provides a way to mold hydrogels and the morphology of muscle tissue and will be beneficial for addressing the need for compliant and optically clear substrates in modeling the neuromuscular junction.

Turning and Turbulence: A Comparative Study of Agility and Fluid Mechanics in Men’s and Women’s Soccer

Thu, 01 May 2025 00:00:00 GMT

Turning and Turbulence: A Comparative Study of Agility and Fluid Mechanics in Men’s and Women’s Soccer Sonner, Jessica E. Female soccer players demonstrate high levels of agility but remain underrepresented in research and experience anterior cruciate ligament (ACL) tears two to eight times more frequently than their male counterparts [1]. These injuries are often associated with high-torsion movements at the knee, such as quick change-of-direction maneuvers in soccer [2]. To examine gender-based differences in agility, this study introduces an in-game metric based on change-of-direction speeds, derived from center-ofmass tracking data from the 2022 Men’s and 2023 Women’s FIFA World Cups. Results show that across positions, ball proximity, and game segments, female athletes tend to change direction both faster and more frequently than male athletes—supporting current injury hypotheses and informing gender-specific cleat design considerations. Beyond individual movement, this study also examines collective team behavior through a fluid mechanics lens. No significant gender differences were found in power spectral densities or second-order structure functions, suggesting symmetry in the underlying coordination dynamics. A direct cascade was observed in the 0–15m range, indicating a consistent transfer of energy across spatial scales. Team dispersion and the Area-Dominant Spread Index correlated with structure function slopes, bridging spatial metrics with turbulence-based models of group behavior.

Deciphering Features of Protective or Maladaptive Cellular Immunity in the Airways Following Primary and Repeated Pathogen Exposure

Thu, 01 May 2025 00:00:00 GMT

Deciphering Features of Protective or Maladaptive Cellular Immunity in the Airways Following Primary and Repeated Pathogen Exposure Bromley, Joshua David The human respiratory tract is constantly subject to environmental stressors and perturbations that cause deviations from homeostatic conditions. The airway’s cellular constituents – epithelial, stromal, and immune cells – maintain local and global homeostasis by facilitating gas exchange and providing a barrier against noxious environmental agents (e.g., xenobiotics, allergens, toxins, and microbes). Infection with viral, microbial, and eukaryotic pathogens can disrupt airway homeostasis, leading to local and systemic inflammation, which can either contribute to the clearance or persistence of the pathogen. Prior antigenic exposure - prophylactically or from a previous infection - can promote transient and long-lived changes in cellular epigenetics, gene expression networks, and cell type composition that may contribute to protective (or maladaptive) immunity; however, we lack a complete understanding of the pathogen and cellular determinants that modulate immunity upon reinfection. In this thesis, we employed single-cell RNA-seq (scRNA-seq), computational methods, and microbial assays to discover the host and pathogen determinants governing airway homeostasis during primary infection and reinfection at barrier sites where the infection begins and may persist: the nasopharynx, airways, and lung parenchyma. First, we leveraged scRNA-seq to identify the cellular and molecular features of mild, moderate, and severe COVID-19, revealing that persons with severe COVID-19 have blunted anti-viral immunity in the nasopharynx. We further extended these findings by profiling nasopharyngeal swabs from vaccinated and unvaccinated individuals across three waves of SARS-CoV-2 variants, revealing shifts in viral tropism and that intramuscular COVID-19 vaccines promote the recruitment of putative antigen presenting macrophages to the nasal mucosa. Next, we used rhesus macaques to interrogate temporal host-pathogen interactions during SARS-CoV-2 infection and reinfection in the lower respiratory tract. This work identified innate training-like gene programs among myeloid populations that provided enhanced protection against SARS-CoV-2 reinfection. Finally, we used cynomolgus macaques as a model to study Mtb infection and reinfection, demonstrating that CD4+ T cells are required to restrict bacterial growth and induce protective immunomodulatory gene programming and cell-cell interaction networks in pulmonary granulomas formed following Mtb reinfection. These findings extend beyond long-held paradigms of protective TB immunity, revealing that CD4+ T cells regulate pro- and anti-inflammatory granuloma equilibria. Collectively, the work presented in this thesis highlights the utility of single-cell genomics for studying respiratory infection- and immuno-biology and provides a framework for contextualizing pathogen-induced deviations from biological homeostasis in the airways, which has implications for the development of prophylactics and therapeutics.

Expansion Microscopy of Extracellular Space for Light Microscopy-Based Connectomic Analysis

Thu, 01 May 2025 00:00:00 GMT

Expansion Microscopy of Extracellular Space for Light Microscopy-Based Connectomic Analysis Emenari, Amauche In this dissertation, we present an exploratory methodology, termed expansion microscopy of extracellular space (ExECS), designed to enhance the visualization of the extracellular space (ECS) within aldehyde-fixed tissue. This technique leverages the principles of expansion microscopy (ExM), a method that facilitates nanoscale imaging on conventional microscopes through physical magnification of specimens, thereby supporting improved visualization of various cellular and tissue components including proteins, nucleic acids, and lipids 1. The ECS forms a continuous environment between cells2. Its presence throughout neural tissue makes it an attractive target for contrast-based techniques such as shadow imaging, where the ECS is selectively labeled to produce negative contrast, revealing cell shapes and boundaries as unlabeled silhouettes within a labeled background. Although ECS delineation in fixed tissue is limited by the fidelity of fixation and may not fully reflect its live-state structure, the resulting contrast with the intracellular environment may offer useful contrast for investigating neural morphology and connectivity, offering a useful approximation of network organization. A key component of the ExECS methodology is the introduction of a customengineered ECS Filler solution. This formulation, detailed later, includes a macromolecular probe intended to serve as a proxy for the ECS. When applied to aldehyde-fixed tissue, the filler is designed to diffuse throughout the sample, preferentially occupying extracellular compartments while remaining largely excluded from intracellular regions. This selective distribution is expected to persist even in areas where aldehyde fixation may have increased membrane permeability. This diffusion behavior is presumed to result from a combination of size-based exclusion and intermolecular interactions between the hyaluronan polymers, which form the main component of the filler solution, and the plasma membrane. The constituent hyaluronan is functionalized with amine groups to enable covalent crosslinking and with azide groups to allow fluorescent tagging via click chemistry. These modifications are intended to enable the ECS filler to act as a contrast agent by labeling the extracellular space, providing a foundation for a shadow-based imaging strategy to delineate morphology of cellular structures. In parallel, we introduce a lipid-targeted form of ExM, termed membrane expansion microscopy (mExM). This approach employs a custom chemical tag that enables nanoscale optical imaging of lipid membranes using a lipid-optimized expansion protocol. mExM, via a novel post-expansion antibody labeling protocol, enables protein-lipid relationships to be imaged in intracellular organelles. This technique may offer new opportunities to examine aspects of neural circuitry by linking cellular morphology with molecular identity. Together, ExECS and mExM offer a potential basis for a light microscopy-based framework for connectomic reconstructions. Unlike traditional electron microscopy approaches, which are labor-intensive and low-throughput3, this strategy aims to improve throughput in mapping of neuronal morphology with enhanced resolution that surpasses diffraction limitations. With the aim of bridging the gap between tissue ultrastructure and optical accessibility, this work may contribute to efforts toward scalable, high-resolution analysis of neural tissue organization.

Decoding Disease Drivers Through Single-Cell Omics and Scalable Phenotypic Screens

Thu, 01 May 2025 00:00:00 GMT

Decoding Disease Drivers Through Single-Cell Omics and Scalable Phenotypic Screens Liu, Nuo At the heart of any human disease is an imbalance between normal and aberrant physiological processes— a disproportion between hypo-immunity and hyperimmunity—a lack of homeostasis. In many cases, a more comprehensive understanding of the molecular basis underlying disease progression and therapeutic failure is still required to devise new strategies for improving patient outcomes. Technological advancements in biomedical research, especially in single-cell omics (e.g. single-cell RNA sequencing, single-cell spatial profiling) have given us unprecedented power to decipher the intricate cellular and molecular features that maintain—or disrupt—this balance. However, validating the causality of these features remains a huge challenge, as the wealth of data often results in a considerable number of hypotheses to test. In this thesis, I explore applications of single-cell genomics tools to understand cellular features associated with disease, with a particular focus on tuberculosis (TB). I then present a potential solution for performing phenotypic screens at scale. In the first part, I applied single-cell RNA sequencing and analysis to human lung samples from a TB-endemic region in South Africa. Using contrastive analysis, I identify key cell populations that are differentially abundant between TB-diseased and TB-negative lung including several neutrophils, macrophages, and fibroblasts subsets. I discovered a de novo gene program highly enriched in the MMP1+CXCL5+ Fibroblast that correlates with TB burden in a non-human primates (NHP) granuloma dataset, supporting the importance of this subset in TB. In a collaborative effort, we validate that this MMP1+CXCL5+ Fibroblast localizes to TB granuloma on independent TB-diseased lung tissues using immunohistochemistry assays and recapitulate the induction of this population from lung-derived fibroblast through in vitro stimulation experiment with M.tb. I further report a SPP1+ macrophage population that is enhanced in TB diseased lungs through single-cell analysis. Moreover, I identified a prominent cross talk between SPP1+ macrophages and fibroblasts in TB diseased lung that mimics similar observations in cancer and fibrosis, supporting an important role for this axis in TB. These distinctive cell populations could serve as potential targets for novel host-directed therapies in tuberculosis. In the second part, I developed a method to compress small molecule phenotypic screens by designing randomized drug pools with replicates of distinct candidates across different drug pools. Our team demonstrated that linear regression models can be applied to computationally deconvolute the individual hits, enabling the identification of top effectors for downstream validation. We benchmarked and demonstrated the efficacy of this approach in a cost-effective imaging platform and then moved into applications on pancreatic ductal adenocarcinoma (PDAC), where we discovered a new perturbation response signature to IL-4/IL-13 with prognostic value for patient survival. We also showcased the utility of this tool on understanding immunomodulation effects in heterogenous mixtures of primary blood cells. Together, this thesis describes novel cellular features important to TB in human lungs, offering new insights that complement existing knowledge from animal models. It also presents a bold, yet effective strategy to scale up phenotypic screen across different biological systems, providing a much-needed solution that bridges the translational gap between human disease and experimental model.

Leaky Vessels

Thu, 01 May 2025 00:00:00 GMT

Leaky Vessels Cong, Frank (Haotian) This thesis serves as a written synthesis of my art practice. It starts with Louis Pasteur’s swan neck flask, Robert Boyle’s air pump, the theater of proof, and cabinets of natural historians to discuss the intentional gesture of containment, exclusion, and controlled permeability in scientific containers and the knowledge production paradigm behind them. I argue that these containers possess another intrinsic gesture – to leak – that opens space for social and cultural dimensions to engage. I propose “leaky vessels” as an analytical tool and a methodology that foregrounds the tension between intentional and unintentional in order to attend to the issues of care, belief, and labor that arise within this dynamic. Chapter 2 develops the concept of “leaky” in three aspects – aesthetic intervention, historical residue, institutional sabotage – by analyzing art practices by Eve Andrée Laramée, Oron Catts and Ionat Zurr, Candice Lin, Maria Thereza Alves, Critical Art Ensemble, and Claire Pentecost. Each case demonstrates how alternative approaches to apparatuses can expose and unsettle the systems of control that govern knowledge authority, allowing seepage, contamination, and embodied histories to return to spaces designed to exclude them. Chapters 3 and 4 turn inward to examine my own art practice, Guardian and The Guarded (2024), RapidRise (2024), and Sweat Dough (2025). In Chapter 3, I discuss the experience of entering the biomaker space at MIT and cultivating animal cells in a pendant, interrogating how care, proximity, and cosmology might challenge the lab’s sterile and utilitarian logic. Chapter 4 discusses the other two projects that operate outside the lab, where I investigate how bodily entanglement with dough fermentation can leak into the broader context of food cultures, labor histories, and symbolic inheritance. Together, these chapters propose a practice that embraces contamination and relationality. Those that leak in and leak out are precisely where new layers of meaning reside.

Data-Driven and Dynamically Feasible Trajectory Generation for Real-Time Powered Descent Guidance and Robotic Exploration

Sun, 01 Sep 2024 00:00:00 GMT

Data-Driven and Dynamically Feasible Trajectory Generation for Real-Time Powered Descent Guidance and Robotic Exploration Briden, Julia Increasingly complex and high-mass planetary missions require autonomous long-horizon trajectory generation to achieve dynamically feasible powered descent guidance. While analytical and indirect methods are computationally efficient, significant simplifications of the dynamics and constraints are required for both problem formulations. Numerical optimization algorithms enable minimum-energy trajectory generation subject to system dynamics and safety constraints but currently remain computationally infeasible on flight-grade processors, taking seconds to minutes to compute a single trajectory. The objective of this dissertation is to develop new algorithms to advance the state of the art in trajectory optimization and planning for autonomous systems. Due to the limited computational abilities of radiation-hardened processors and an increased need for spacecraft and robotic autonomy, specialized algorithms capable of running in realtime constitute enabling technologies for space exploration. Three major contributions are developed in this dissertation. First, a transformer neural network-based algorithm is created to predict the tight constraints that recover the solution and parameter sets for constrained optimization problems. By training on prior runs of the numerical optimization solver, the learned mapping can construct a reduced problem formulation that recovers the optimal solution while reducing runtime by up to an order of magnitude. Second, a method to embed problem-specific information into the neural network training process was developed. By embedding the Lagrangian and Lagrangian gradient merit functions into the training process, neural network-generated control policies are biased toward constraint satisfaction. Third, an autonomous hybrid targeting and guidance algorithm was designed to utilize probabilistic risk maps and numerical optimization to select and navigate to minimum-risk landing sites. Applications in planetary powered descent and landing, as well as rover path planning, are used to benchmark algorithm performance.

The simulation of a multi-product, multi-department factory

Wed, 01 Jan 1964 00:00:00 GMT

The simulation of a multi-product, multi-department factory Levy, Donald Stephen. Thesis: B.S., Massachusetts Institute of Technology, School of Industrial Management, 1964

Comparative elastic and plastic analysis and design of steel frames

Fri, 01 Jan 1960 00:00:00 GMT

Comparative elastic and plastic analysis and design of steel frames Padilla Valenzuela, Rodolfo Augusto. Thesis: B.S., Massachusetts Institute of Technology, Department of Civil and Sanitary Engineering, 1960

Dynamics of torpedo depth control

Wed, 01 Jan 1992 00:00:00 GMT

Dynamics of torpedo depth control Carleton, John Thomas. Thesis: M.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1992; Includes bibliographical references (leaf 72).

Quantum theory of mode locking.

Fri, 01 Jan 1971 00:00:00 GMT

Quantum theory of mode locking. Lang, W. R. (W. Roy) Thesis: Ph. D., Massachusetts Institute of Technology, Department of Physics, 1971; Vita.; Bibliography: leaves 88-90.

An investigation of the engineering aspects of a wind tunnel magnetic suspension system

Thu, 01 Jan 1959 00:00:00 GMT

An investigation of the engineering aspects of a wind tunnel magnetic suspension system Chrisinger, John Edvil. Thesis: M.S., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 1959; Includes bibliographical references (leaf 62).

Extrageniculate and extrastriate affiliates of the geniculocortical pathway in the cat

Tue, 01 Jan 1980 00:00:00 GMT

Extrageniculate and extrastriate affiliates of the geniculocortical pathway in the cat Berson, David Matthew. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Psychology, 1980; Vita.; Bibliography: leaves 114-126.

New theoretical methods for the study of the electronic structure of solids.

Sun, 01 Jan 1978 00:00:00 GMT

New theoretical methods for the study of the electronic structure of solids. Mele, Eugene John. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Physics, 1978; Includes bibliographical references.

A financial history of the Boston elevated

Fri, 01 Jan 1926 00:00:00 GMT

A financial history of the Boston elevated Stallman, Edward B.; Bush, Horace McM. Thesis: B.S., Massachusetts Institute of Technology, Department of Business and Engineering Administration, 1926; Includes bibliographical references (leaf 34).

Design and Evaluation of Skill-Based Imitation Learning Policies for Robotic Manipulation

Thu, 01 May 2025 00:00:00 GMT

Design and Evaluation of Skill-Based Imitation Learning Policies for Robotic Manipulation Palleiko, Andrew Imitation learning is a popular approach for obtaining intelligent robotic policies by learning from human demonstrations. Within this field, there is significant interest in the development of multi-task architectures that can efficiently learn diverse sets of tasks. Skill-based imitation learning methods, which abstract action sequences into ``skill'' representations for planning, offer structural advantages for handling the challenges of multi-task imitation learning that make them an attractive option for this problem. This work presents a novel skill-based imitation learning architecture formulation, with a causal transformer VAE skill-abstraction network paired with an autoregressive transformer planning policy. We find that our skill-abstraction network shows promise in identifying meaningful skills, but that the chosen planning architecture is poorly suited for predicting these skills due to multimodality in the resulting latent space. This is followed by a set of evaluations applied to an existing skill-based method with comparisons to a non-skill-based network on a multi-task dataset. We systematically investigate the performance impacts of six different policy and dataset conditions: data quantity, task variety, retry behavior, control precision, goal representations, and zero-shot transfer. Our experiments reveal limited increases in skill-based policy performance with more demonstrations or task variety, but improvements across architectures through exposure to demonstration retry behavior. Overall, the skill-based architecture demonstrates superior robustness to goal representation variations and low-level process noise than the non-skill-based policy, while neither architecture achieves meaningful zero-shot generalization to novel task combinations. These findings provide insights into the current state of IL methods, with the additional goal of establishing a framework for the evaluation of future multi-task IL architectures.

Towards Multimodal Streaming Perception: A Real-Time Perception Scheduling Framework Based on Relevance

Thu, 01 May 2025 00:00:00 GMT

Towards Multimodal Streaming Perception: A Real-Time Perception Scheduling Framework Based on Relevance Huang, Dingcheng In modern human-robot collaboration (HRC) applications, multiple perception modules jointly extract visual, auditory, and contextual cues to achieve comprehensive scene understanding, enabling the robot to provide appropriate assistance to human agents intelligently. While executing multiple perception modules on a frame-by-frame basis enhances perception quality and information gains in offline settings, it inevitably accumulates latency, leading to a substantial decline in system performance in streaming perception scenarios. Recent work in scene understanding, termed Relevance, has established a solid foundation for developing efficient methodologies in HRC. However, modern perception pipelines still face challenges related to information redundancy and suboptimal allocation of computational resources. Drawing inspiration from the relevance concept and the inherent sparsity of information in HRC events, we propose a novel lightweight perception scheduling framework that efficiently leverages output from previous frames to estimate and schedule necessary perception modules in real-time. Our experimental results demonstrate that the proposed perception scheduling framework effectively reduces computational latency by up to 27.52% compared to conventional parallel perception pipelines, while also achieving a 72.73% improvement in MMPose accuracy and comparable YOLO accuracy. Additionally, the framework demonstrates high keyframe accuracy, achieving rates of up to 98% in dynamic scenes. The results validate the framework’s capability to enhance real-time perception efficiency without significantly compromising accuracy. Additionally, the framework shows potential as a scalable and systematic solution for multi-modal streaming perception systems in human-robot collaboration.

Fracture Mechanics of Networks

Thu, 01 May 2025 00:00:00 GMT

Fracture Mechanics of Networks Hartquist, Chase M. Networks of interconnected materials permeate throughout nature, biology, and technology due to exceptional mechanical performance. Despite the importance of failure resistance in network design and utility, no existing physical model effectively reconciles strand mechanics and connectivity to predict fracture in diverse networks that constitute polymeric, architected, and biological materials. While traditional models predict the intrinsic fracture energy – the minimum energy to propagate a crack per unit area – of a polymer network is the energy to rupture a layer of chains, they can underestimate experiments by up to two orders of magnitude. In Part I, we show that the intrinsic fracture energy of polymer-like networks stems from nonlocal energy dissipation. We then reveal a general scaling law that captures nonlocal energetic contributions and connects strand mechanics with topological connectivity to universally predict the intrinsic fracture energy of stretchable networks. We measure intrinsic fracture energy using experiments and simulations of 2D and 3D networks with various strand constitutive behaviors, defect densities, strand length distributions, lattice topologies, and length scales. Results show that local strand rupture and nonlocal energy release contribute synergistically to the measured intrinsic fracture energy in networks. These effects align such that the intrinsic fracture energy scales independent of the energy to rupture a strand; it instead depends on the strand rupture force, breaking length, and connectivity. In Part II, we present a model for real polymer fracture and design elastomers with highly regular connectivity. End-linking then deswelling star polymers produces a class of elastomers with low defects and no trapped entanglements, enabling ultrahigh straininduced crystallinity of up to 50% and stretchability that scales beyond the saturated limit. These features promote a pronounced elastocaloric cooling effect and enable reversible two-way tuning of thermal conductivity by strain or temperature modulation. The mechanical and thermal properties of these polymer networks offer promise in addressing challenges in clean energy, thermal management, and biomedicine. Our findings establish a physical basis for understanding network fracture and design principles for fabricating tough polymeric, biological, and architected materials across multiple length scales for advanced applications.

Modeling the Sit-to-Stand Transition using Koopman Lifting Linearization and Human State Estimation

Thu, 01 May 2025 00:00:00 GMT

Modeling the Sit-to-Stand Transition using Koopman Lifting Linearization and Human State Estimation Bell IV, John H. The Sit-to-Stand (STS) transition is one of the most dangerous daily activities for the elderly population, as it is one of the situations in which falls occur most often. Despite its risks, STS dynamics remain poorly understood, and current STS assistance devices fail to utilize knowledge of STS dynamics to effect their support. This thesis presents contributions to the dynamic modeling of STS and to human-robot collaboration for improving robotic assistance of STS. To coherently capture the multi-phase nature of STS, lifting linearization, a dynamic modeling methodology inspired by Koopman operator theory, to subsume segmented local dynamics in a globally linear dynamic model. A novel class of lifting linearization basis functions, termed “State-Membership Product (SMP)” observables, enables both the seamless blending of local dynamics into a global model, and the direct extraction of phase-specific behaviors from the global model. It is shown that an SMP-Koopman linear model tuned to published data of STS experiments is capable of reproducing the multi-phase STS dynamics with a single linear model. Building on this framework, STS is additionally modeled as a lifted linear feedback control system, composed of an SMP-Koopman-based open-loop biomechanical model of the human body and a linear quadratic regulator (LQR) which guides the body to stand up. The LQR controller, tuned to replicate STS motion, guides the human body model through the phases of STS without explicit phase-switches, improving system robustness. To enhance human-robot collaboration in STS assistance, a framework for estimating patient cooperativeness is also introduced, leveraging a simplified dynamic model and an Extended Kalman Filter. By analyzing a human’s initial response to applied physical and verbal cues, the estimation framework assesses willingness to engage in assisted STS. Together, these contributions advance both the modeling and estimation of STS, offering insights crucial for the development of safe, effective robotic assistance.

Verse and Reversal: The Poetic Return to the Inner Child as Black Revolutionary Praxis

Thu, 01 May 2025 00:00:00 GMT

Verse and Reversal: The Poetic Return to the Inner Child as Black Revolutionary Praxis Dunnell, Kaelyn Black revolutionary movements historically have centered the role of the Black child—as either foundation, visionary, or representation of Black liberation. The identity of any given revolutionary movement is characterized by three tenets: resistance, imagination, and love. In order for the individual to uncover the origin of these three tenets for themselves. This thesis is about the poiesis of the revolutionary—the making and re-making of the revolutionary—and in it I argue that the very process of forming revolutionary identity is poetic. I coin the phrase poetic revolutionary to capture that process, which involves tapping into the font of revolutionary soulfulness, which is one’s inner child or the voice and experience of the Black child. The literature guiding this analysis is from June Jordan’s archive hosted at Schlesinger Library, with Voice of the Children, a children’s publication edited by Jordan, as one of the most notable works. I examine June Jordan as the model of the Black revolutionary who has uncovered the language of her child, and I also examine the works of the children she worked with (whose 13– 15-year age ranges, notably, are on the cusp of the definition of childhood that I adopt in this thesis—more in Section I). I gather evidence from workshop diary entries written by Jordan and by her students, poetry excerpts from Voice of the Children, and Jordan’s own writing from her childhood and beyond to support my theory of the poetic revolutionary.

Quantifying Human Balance Performance and Control to Inform Therapy

Thu, 01 May 2025 00:00:00 GMT

Quantifying Human Balance Performance and Control to Inform Therapy Shiozawa, Kaymie S. Maintaining balance is essential for daily activities and overall health. However, balance capability often declines with age or due to health conditions such as stroke, increasing fall risk. Falls among older adults are a major public health concern, affecting 14 million older adults annually in the US and directly causing over 40,000 deaths. Timely and accurate assessment of balance impairment is crucial to prevent falls and promote independence. Current assessments rely heavily on subjective therapist evaluations, underscoring the need for objective, quantitative methods. With the growing strain on healthcare systems due to an aging population, continuous at-home balance monitoring is also increasingly important. Additionally, a comprehensive understanding of the motor control mechanisms that deteriorate with aging or disease is crucial for informing therapy methods and technologies. The goal of this thesis was to develop and validate methods that quantify quiet balance ability and control in unimpaired and impaired human participants. The first part focuses on assessing balance ability, the capacity to maintain upright posture during quiet stance that is currently often quantified by measures of body sway. A review of the strengths and limitations of current clinical and instrumented balance assessments highlighted a critical need for continuous assessment methods that enable objective monitoring of balance function outside of clinical settings. Addressing this need, a novel algorithm that quantifies balance ability using only force and motion sensors embedded in an instrumented cane was developed. Well-established balance measures were successfully estimated in both younger and older adults, demonstrating the proposed method's potential to facilitate continuous balance monitoring in real-world environments. The next part focuses on identifying balance control strategies. The novel intersection-point analysis, based on foot-force direction and point of application, was used in conjunction with a simple biomechanical model and an optimal controller to quantify balance control. The first study demonstrated that unimpaired quiet balance in a challenging environment was best described by a controller that maintained minimal effort by adjusting relative ankle and hip joint torques. Applying this method to aging populations in a subsequent study revealed that older adults rely more on neural feedback, possibly to compensate for muscle strength deficiency. This study also quantified individual balance controllers, highlighting the method's potential as a diagnostic tool for aging populations. Finally, the model was extended to describe balance control after stroke. The results suggest that the non-paretic limb compensated for the paretic limb's abnormal coordination pattern by strongly favoring neural feedback. As one of the first studies to model quiet balance after stroke, this work lays the foundation for future efforts on studying balance impairments. The contributions of this thesis are instrumental to enhancing at-home monitoring, advancing clinical practices, and reducing fall-related injuries, ultimately improving quality of life for aging and neurologically impaired populations.

Deformable Object Manipulation with a Tactile Reactive Gripper

Thu, 01 May 2025 00:00:00 GMT

Deformable Object Manipulation with a Tactile Reactive Gripper Sunil, Neha Manipulating deformable objects remains a fundamental challenge in robotics, as techniques developed for rigid objects often fail to generalize. Deformable objects exhibit infinite-dimensional configuration spaces, frequent self-occlusion, and high model uncertainty, making global state estimation and predictive modeling unreliable. To address these challenges, we propose a perception-driven framework that combines global visual understanding with local tactile feedback. Rather than modeling the full configuration of the object, we leverage local constraints, grounded in modular visual and tactile representations, to enable robust, reactive, and generalizable manipulation. The primary contributions of this work include: • Chapter 2: Cable Following. A tactile control strategy for in-hand cable manipulation that decouples contact regulation from object pose control, enabling fast, reactive sliding and closed-loop plug insertion using only local tactile feedback. • Chapter 3: Towel Edge Tracing. An extension of contact-based control to fabric edge following and the learned tactile perception networks to support this capability. • Chapter 4: Visuotactile Grasp Affordance. A grasp affordance model trained in simulation and refined with tactile self-supervision, enabling high-confidence edge grasping on towels. • Chapter 5: Dense Object Correspondence. A confidence-aware dense descriptor representation. Supports correspondence across crumpled and symmetric garments in air and on a table. • Chapter 6: Behavior Architecture and Planning Interfaces. Integration of perception modules into a reactive, confidence-based folding system and an exploration of how dense descriptors can interface with demonstrations, language, and task and motion planning. Collectively, these contributions show that global state estimation and dynamics prediction are not required for reliable deformable manipulation. Instead, semantically meaningful local interactions, guided by modular visual and tactile representations, can drive scalable, long-horizon behaviors across varied objects, configurations, and tasks.

Fluid Sealing Challenges in Solid Oxide Electrolysis Cells and Rapid Swap Battery Systems

Thu, 01 May 2025 00:00:00 GMT

Fluid Sealing Challenges in Solid Oxide Electrolysis Cells and Rapid Swap Battery Systems Lindberg, Ian G. This thesis explores the design and development of several mechanical elements relevant to two technologies Important to a global transition to green energy, hydrogen and electric vehicles. The portion of the thesis relating to hydrogen focuses on preloading mechanisms and high temperature seals, two design spaces crucial to the implementation of solid oxide hydrogen generation. Due to the high operating temperatures (600°C - 800°C), seal materials commonly used in other applications are inadequate and glass or vermiculite based seals must be used. The delicateness of these seals makes them a common failure point, and consistent application of a preloading force is key to mitigating this. The concept of a variable-bypass piston is proposed as a preloading mechanism suitable for the high temperatures present inside solid oxide electrolyzer systems, and the development of seal geometries as well as flow characterization of porous steel wool seals to enable parametric design is documented. As an alternative to current sealing methods, initial development of a composite seal utilizing materials and manufacturing methods originating in the semiconductor industry was also conducted. The final section of the thesis proposes the concept and covers initial testing of fluid transfer through a kinematic coupling, a topic of potential interest for implementing liquid pack cooling in a system of rapidly swappable batteries for electric vehicles.

Enhancing the Performance of Skeletal Muscle Powered Biohybrid Robots

Thu, 01 May 2025 00:00:00 GMT

Enhancing the Performance of Skeletal Muscle Powered Biohybrid Robots Bawa, Maheera Skeletal muscle powers all voluntary motion in many living creatures, enabling behaviors such as walking, jumping, swimming, and flying. The field of biohybrid robotics aims to use biological actuators, such as skeletal muscle, to power adaptable robots that respond to their environment. Previous work in this field has focused on deploying 3D skeletal muscle tissues to power robotic function. In natural systems, muscles can also be organized in 2D formats to power a range of movements such as fish-like swimming and peristaltic pumping. However, long-lasting 2D cultures of skeletal muscle have been precluded by force-generating cells delaminating from their underlying substrate. Building on previous work from our lab demonstrating a method to culture contractile skeletal muscle in 2D formats, this work aims to enhance the performance of these systems by tuning substrate stiffness and topography. We show that optimizing system parameters prolongs actuator lifetime and enhances force by 100x.

Development and Implementation of a Smart Factory for Educational Fiber Extrusion Device Production

Thu, 01 May 2025 00:00:00 GMT

Development and Implementation of a Smart Factory for Educational Fiber Extrusion Device Production Fillon, Marie This thesis presents the development and production of FrED (Fiber Extrusion Device), an educational manufacturing system designed to bridge the gap between theoretical instruction and hands-on practice in process control, computer vision, and smart manufacturing. Building on an existing prototype, this work focused on transitioning FrED from a proof-of-concept into a production-ready system by designing scalable workflows, improving hardware and software integration, and developing tools to ensure traceability and repeatability across builds. A major contribution of this thesis was the enhancement and implementation of a smart factory environment capable of supporting batch production. This included designing and deploying applications using Tulip Interfaces to manage inventory, guide subassembly processes, and monitor production metrics in real time. A modular SKU system and structured bin labeling framework were introduced to reduce errors, maintain version control, and support future growth. Station-specific apps were developed and refined to ensure consistent assembly and simplify onboarding across a rotating team of users. In parallel, this thesis contributed to the evaluation and refinement of a vision-based diameter measurement system using a low-cost USB camera. The system was analyzed under various operating conditions and its limitations under motion and variable lighting were quantified. Multiple image processing strategies were explored and robustness metrics were developed to inform future improvements. To ensure pedagogical relevance, the system was tested in user-facing workshops and public demo sessions. Feedback informed updates to both the assembly process and instructional content. By the end of the development cycle, the system supported the successful production of 35 complete FrED units, establishing a replicable model for small-scale manufacturing. This thesis demonstrates how modular digital infrastructure can enable scalable hardware deployment. It also highlights the practical challenges of transitioning from prototype to production and proposes tools and methods that can support broader adoption of smart manufacturing principles in learning environments.

Analyzing Vibration in Omni-Wheels: A Design of Experiments Approach to Optimizing Omni-Wheel Selection

Thu, 01 May 2025 00:00:00 GMT

Analyzing Vibration in Omni-Wheels: A Design of Experiments Approach to Optimizing Omni-Wheel Selection Sanghai, Rohan S. Omni-wheels, known for enabling holonomic motion in robotic systems, often introduce vibration due to their complex geometry and multiple contact points. Unlike caster wheels with established testing standards, omni-wheels lack comprehensive characterization methods. While parallel studies by Ilkbahar [1] and Donnellan [2] explore their rolling resistance and static load capacity, a systematic analysis of vibration characteristics remains absent from the literature. This thesis presents an investigation of the vibration behavior of various omniwheel designs using a Design of Experiments (DOE) approach. A full factorial experimental design was developed, considering factors such as wheel type, rotational speed, applied load, and wheel orientation angle. Individual regression models were developed for each of six wheel types, treating operational parameters as continuous variables. Vibration levels were measured using root mean square (RMS) acceleration, derived from Fast Fourier Transform (FFT) and Power Spectral Density (PSD) analyses of accelerometer data. Results show that rotational speed consistently increased vibration across all wheel designs, while lateral motion (90° angle) consistently reduced vibration compared to forward motion. The effect of applied load varied significantly between wheel designs, with some wheels showing reduced vibration under load while others remained unaffected. Wheels DZ(1) and Vex(5) demonstrated the lowest average vibration levels, though post-test inspection revealed trade-offs with durability, including roller deformation and material degradation. Interaction effects, particularly between angle and speed, were statistically significant for all wheel types, indicating that the benefits of lateral motion are enhanced at higher speeds. This research provides a framework for optimizing omni-wheel selection to minimize vibration by developing wheel-specific predictive models that quantify sensitivities and interaction effects across various designs and conditions, improving system performance and stability. The findings highlight that wheel selection must consider not only vibration performance but also trade-offs with durability and rolling resistance, establishing vibration characteristics as a critical consideration alongside other performance metrics when selecting omni-wheels.

A Model-Based Planning and Control Framework for Parkour-Style Legged Locomotion

Thu, 01 May 2025 00:00:00 GMT

A Model-Based Planning and Control Framework for Parkour-Style Legged Locomotion Chignoli, Matthew T. Legged robots have long been envisioned as a means of expanding robotic capabilities beyond structured environments, yet achieving high-agility locomotion remains a fundamental challenge. This thesis presents a model-based framework for parkour-style locomotion, enabling robots to execute highly dynamic maneuvers such as jumps, rolls, and flips with precision and robustness. A key challenge in planning these motions is selecting an appropriate dynamic model that balances computational efficiency with physical accuracy. To address this, a model assessment strategy is introduced to determine the simplest model capable of capturing task-relevant dynamics. Even with well-chosen models, solving long-horizon trajectory optimization problems for dynamic motions is computationally demanding. This thesis introduces graduated optimization techniques, which improve solver efficiency and reliability by generating high-quality initial guesses through progressively refined problem formulations. Additionally, a novel formulation of rigid-body dynamics algorithms for systems with kinematic loops accelerates trajectory optimization and simulation. Finally, two control strategies are proposed to execute planned motions on hardware: a model-based tracking controller for real-time adjustments and an imitation learning policy trained on optimal trajectories to enhance robustness. Extensive experiments on hardware validate the framework, demonstrating the successful execution of complex, high-impact locomotion behaviors. By integrating advanced planning, optimization, and control techniques, this work establishes a foundation for high-agility legged locomotion, pushing beyond conventional automation toward real-world, dynamic robotic movement.

Designing and Optimizing Magnetohydrodynamic Induction Marine Energy Harvester

Thu, 01 May 2025 00:00:00 GMT

Designing and Optimizing Magnetohydrodynamic Induction Marine Energy Harvester Scali, William T. Magnetohydrodynamic (MHD) power generation presents a promising approach for harvesting energy from marine environments, offering a sustainable alternative for powering naval assets and coastal infrastructure. While energy harvesting technologies are widely used in terrestrial and aerial applications, their implementation in marine environments remains limited. This thesis explores the feasibility of an MHD Inductive Marine Energy Harvester, optimizing its design for undersea naval applications to enhance energy efficiency and reduce carbon emissions with minimized construction costs. A theoretical 2D model was developed based on Maxwell’s equations and Fourier analysis to characterize the physics governing MHD power generation in seawater. This model was extended to multiple concentric gaps on one device, refining predictions of power output under varying flow regimes. Numerical simulations using MATLAB enabled the evaluation of key parameters, including fluid conductivity, magnetic field strength, and shroud design, to optimize energy conversion efficiency. Furthermore, geographical and coastal tide analyses were conducted to determine optimal deployment locations, maximizing power extraction from natural marine currents. Economic viability was assessed through a cost-benefit analysis, comparing the energy yield per unit cost of the harvester against existing renewable energy technologies and other maritime power sources. Results indicate that under specific conditions, MHD generators can effectively supplement energy demands, reducing reliance on conventional fuel or other electrical power sources. The findings of this research contribute to the advancement of marine renewable energy technologies, demonstrating the potential of MHD induction-based harvesting as a scalable solution for sustainable power.

Solar-Powered Critical Cooling: A Theoretical Feasibility Study for Human Thermal Regulation

Thu, 01 May 2025 00:00:00 GMT

Solar-Powered Critical Cooling: A Theoretical Feasibility Study for Human Thermal Regulation Hall, Jeff Over the last 50 years, the leading global environmental hazard has not been hurricanes, lightning, tornadoes, floods, or earthquakes, but extreme heat events. With climate models projecting an increase in the frequency, intensity, and duration of heatwaves in the coming decades this threat to life is expected to only increase. Air conditioning has been demonstrated to reduce mortality during heatwaves yet uses an order of magnitude more energy than necessary to keep a human cool. Using principles of similitude to extrapolate the capability of existing vapor compression equipment, an objective function to maintain energy balance in a human exposed to extreme heat is developed across a design space. The function shows that in a standard forced convection air conditioning system, there no opportunity to provide emergency cooling of a human due to the slow mass flow rate needed to cool air in a single stream. As such, status-quo attempts to cool humans with general-purpose air conditioning will always be an inefficient use of energy. By focusing on keeping people cool, not spaces, we propose three paths forward for critical human cooling that appropriately match the energy needs of humans: radiative cooling, liquid cooling devices, and low-mass flow air conditioning.

Fractured Practices: How Schooling Norms Limit Modeling Practices in Traditional Technical Thermal-Fluids Engineering Courses -- And the Possibilities Emerging through the Cracks

Thu, 01 May 2025 00:00:00 GMT

Fractured Practices: How Schooling Norms Limit Modeling Practices in Traditional Technical Thermal-Fluids Engineering Courses -- And the Possibilities Emerging through the Cracks Huffman, Sandra In professional science and engineering contexts, modeling practices are frequent and diverse. To understand, analyze, and communicate, scientists and engineers simplify and distort the complex systems with which they work. This practice is known as modeling. Typically, scientists create models to predict and explain phenomena while engineers develop them to analyze and test systems, make design decisions, and predict the performance of built systems. Models can include verbal (ex. analogy, story), visual (ex. diagrams, graphs, images), and symbolic (ex. equations) representations. When scientists and engineers model, they do so expansively: pulling from different resources, combining modeling strategies, engaging in critique and iteration, and contextualizing their claims in the work of their field. This is not the case for students in technical engineering classes who are attempting to learn these skills. Traditional, lecture-based courses are the norm for introducing technical material to undergraduate engineering students. These courses typically consist of lectures, recitations, problem sets, and exams. In this type of class, students report homework and test problems as having an outsized influence on their learning approach. These problems tend to be narrow and prescribed. Colloquially known as ‘Textbook-Style’ problems, well-defined, single-solution problems are not sucient to prepare students to successfully tackle the ill-defined, multifaceted engineering problems they will face in their careers. These problems do not elicit student engagement in scientific or engineering modeling practices. Instead, they lead to inauthentic, bounded learning where students develop strategies adequate for groups of similar problems, but too narrow for use outside of the classroom. There has been significant research on innovative educational interventions and alternative problem types shown to improve classroom learning. However, educators work within established structures that resist change, leading to the perpetuation of insucient practices. The gap between textbook-style problems and the problems engineers face, therefore, exists not just in the problem type, but in the context surrounding the task. In this work, I describe and characterize the norms and practices of the classroom environment through three qualitative studies, each centered on traditional technical thermal-fluids courses. Specifically, I investigate the ways in which the development of student modeling practices are supported or undermined. I do this, in part, by adapting the theoretical framework of Figured Worlds. Originally developed by Dorothy Holland and later used in Engineering Education research, figured worlds is a situative framework that allows researchers to look at distinct, sometimes contradictory cultural worlds within the same group and activity. In the first study, I look at individual student approaches to classroom tasks in a think-aloud study, comparing their problem solving approaches and analyzing prompt-student interactions. In the second study, I analyze small groups’ modeling practices and how they are limited by the cultural practices of schooling. In the third study, through semi-structured interviews, I document instructor perceptions of their research and teaching, and discuss the misalignments within and between these contexts. Together, these works outline the mechanisms by which school practices can inhibit the development of student modeling capabilities and the role of students and instructors in perpetuating these practices. In describing student and instructor behavior and contextualizing practices that may otherwise be ascribed to misconceptions, carelessness, or ignorance, I hope to build a foundation for future research into pragmatic educational interventions for enhanced learning outcomes.

Behavioral Methods for Next-Generation Shipboard Power System Simulation: Letting SPARCS Fly

Thu, 01 May 2025 00:00:00 GMT

Behavioral Methods for Next-Generation Shipboard Power System Simulation: Letting SPARCS Fly Almquist, Ethan T. Design requirements on modern naval platforms are increasing the complexity and criticality of onboard electric plants. They form the backbone of warship operational capability and are at the heart of maritime decarbonization. Tasks such as assessing the ship's capacity in a damaged state, optimizing the mission profile of a fleet of vehicles, and evaluating broad design spaces in an efficient manner are increasingly difficult as electric network complexity increases. Traditional modeling techniques are either too computationally expensive, or lack the fidelity necessary to produce meaningful insights into the electric network's operation. Behavioral modeling bridges this gap, but is underdeveloped to support the system architectures of tomorrow's ships. This work details the advancement of behavioral modeling of electrical systems to incorporate hybrid AC/DC and ring bus architectures, the development of parallelization techniques, and SPARCS: a software package offering Shipboard Parallelized Analytics with a Rapid Configuration Simulator.

Design for Longevity: Service and System Innovation

Thu, 01 May 2025 00:00:00 GMT

Design for Longevity: Service and System Innovation Lee, Sheng-Hung The global demographic shift toward an aging population presents complex social, economic, and systemic challenges, necessitating innovative approaches to service design, systems thinking, and financial planning. This dissertation, Design for Longevity: Service and System Innovation, examines these transformations and proposes strategies to foster a “longevity society”, a new era in society necessitating a fundamental rethinking of age and ageing to effectively harness the opportunities afforded by increased life expectancy (Scott, 2021). This research is built upon five relevant paradigm shifts: 1. from age-based to stage-based mindsets, 2. from product-driven to service-driven solutions, 3. from human-centered to humanity-centered design, 4. from circular to longevity economics, and 5. from an aging society to a longevity society. These shifts redefine the role of designers and researchers in creating adaptive, inclusive, and sustainable systems for the future. This dissertation explores how tangible artifacts, Longevity Planning Blocks (LPBs), can be employed to create effective service encounters. The research questions explore 1. how to use boundary objects (BOs) to uncover and define latent user needs, 2. how to use a mixed-method approach to analyze experiment data, 3. data-driven persona creation, and 4. the design of longevity planning services across financial planning, service innovation, and system thinking. Central to the research is a study of LPBs, BOs designed to facilitate collaborative engagement between a facilitator and 69 Boston-based participants, stratified by age, gender, pre-tax annual income, and assets. LPBs, employed in experiments, help investigate participants’ needs and concerns across various life transitions and stages. These tangible BOs facilitated informal yet insightful discussions, uncovering how individuals navigate ambiguity, make complex decisions, manage their evolving physical, mental, and social health, and perceptions about living solo. Data from in-person longevity planning experiments provided nuanced insights into the interplay of individual, societal, and systemic factors shaping longevity planning services. A mixed-methods approach integrates qualitative and quantitative techniques, including expert and user interviews, co-creation workshops, pre- and post-experiment surveys, hierarchical cluster analysis, K-means clustering for persona development, and causal loop diagrams for longevity planning service system modeling. Constructivist grounded theory and exploratory factor analysis uncover emerging themes and systemic interconnections, emphasizing the importance of adaptive services that align with changing needs and broader social infrastructures. The study introduces the notion of Design for Longevity (D4L), expanding on longevity economics and circular economy principles to address the complexities of extended lifespans. D4L highlights how evolving resources, transformative needs, and systems integrate life stages into the design of products, services, and experiences. This dissertation contributes to service innovation, financial planning, and system design by proposing actionable insights for longevity planning services. It emphasizes multi-stage life planning, intergenerational collaboration, and systemic thinking as foundational to a longevity society. This dissertation contributes a mixed-method approach, offering design practitioners a replicable, data-driven framework for persona creation applicable beyond longevity planning. Concluding with reflections on social infrastructure, community, and culture, the study calls for cross-disciplinary collaboration to address longevity planning challenges. By advancing the understanding of longevity planning and its systemic implications, this work lays a foundation for designing a future where extended lifespans are inclusive and socially engaged.

Integrated Prosthetic Leg Design Frameworks for People with an Above-Knee Amputation

Thu, 01 May 2025 00:00:00 GMT

Integrated Prosthetic Leg Design Frameworks for People with an Above-Knee Amputation Petelina, Nina T. A well-fitting, high-performance prosthesis for people with a lower limb amputation can greatly improve users’ mobility and quality of life. Still, many amputees lack access to high-performance prosthetic components due to the cost and availability of continuous care. This thesis aims to design low-cost, high biomechanical performance above-knee prosthetic leg components (prosthetic foot and knee) that will result in a walking motion likely to be perceived as able-bodied after minimal acclimation time. Above-knee amputees have two common gait deviations from able-bodied and below-knee amputee gait: lack of early stance knee flexion (ESF) and delayed initiation of knee flexion (IOF) during late stance phase. These deviations are likely a result of prioritization of stability at the expense of other functions such as shock absorption and progression through stance. A preliminary perception study was conducted to investigate the acceptable bounds of gait deviation that can be incorporated into a prosthetic leg design without compromising the perception of "typical" walking. Using these results, I created the Hip Trajectory Error (HTE) framework for designing prosthetic feet specifically for people with an above-knee amputation. The HTE framework takes into account the lack of ESF by incorporating the shock absorption function of ESF within the prosthetic foot design. This is achieved by targeting able-bodied hip center motion, which is correlated with sufficient shock absorption during the stance phase. This thesis presents an optimization and performance evaluation process that resulted in a prosthetic foot structure that not only closely replicates able-bodied hip center motion but also could be manufactured for a low cost. An experimental study successfully demonstrated that the Hip Trajectory Error (HTE) framework can be used to predictively design prosthetic feet for aboveknee amputees. HTE-designed prosthetic feet enable comparable biomechanical performance to daily-use tuned and prescribed prosthetic feet within 10-15 minutes of acclimation time and without iterative multi-day fittings. Next, I proposed a method to recommend a damping coefficient for the prosthetic knee to achieve able-bodied peak knee flexion during swing phase. A range of recommended damping coefficients to achieve target peak knee flexion angle in transfemoral amputees was determined using a simple three-step framework. This framework incorporates effects from common transfemoral prosthetic gait deviations, such as slower self-selected walking speeds and delay in initiation of knee flexion during late stance. The calculated range of recommended damping coefficients was experimentally investigated and found to enable a peak knee flexion angle within two standard deviations of able-bodied peak knee flexion angle. Lastly, I created the Full Leg Optimization (FLO) framework to design the prosthetic foot and knee concurrently based on minimal inputs from the user and the prosthetist. The framework anticipates the lack of ESF and delay initiation of late stance knee flexion and uses the HTE framework to predict the orientation and location of the knee mechanism. Using this prediction, the rotational axes of the prosthetic knee can be positioned to start knee flexion at a point in late stance chosen by the prosthetist to provide sufficient stability to the user. A proof-of-concept study demonstrated the accuracy of the prediction for one user after minimal acclimation time, confirming the ability to predictively design prosthetic leg components in tandem. The FLO framework can therefore be used to predictively design a passive prosthetic leg for above-knee amputees while considering common gait deviations due to stability needs. This doctoral work demonstrates that the presented frameworks can be used to quantitatively design prosthetic feet and knees based on the needs of above-knee amputees, which could save fitting time, manufacturing cost, and improve mobility.

Multimodal Non-Contact Sensing of Neonatal Vital Signs Using Radar and Video

Thu, 01 May 2025 00:00:00 GMT

Multimodal Non-Contact Sensing of Neonatal Vital Signs Using Radar and Video Chityat, Inbar Preterm neonates represent a vulnerable population which traditional contact-based monitoring devices are not optimized for their small size and complicated physiology. Adhesive sensors and wires can cause infections, discomfort, and impair the delivery of clinical care. Therefore, these most fragile patients could significantly benefit from remote health monitoring. This thesis establishes the foundation for a multimodal device designed for noncontact monitoring of neonates in the Neonatal Intensive Care Unit (NICU) that integrates a video camera and a radar. The device is used to estimate vital signs such as respiratory rate (RR), using both unimodal (solely video or radar) and multimodal fusion approaches that combine data from both sensors. Preliminary testing was conducted on neonatal simulator mannequins, followed by a clinical study at Tufts Medical Center NICU which collected data from 16 neonates so far (with the goal of reaching 20). The collected data was processed, labeled, and organized using image processing techniques and manual review, and then analyzed using a Video Vision Transformer (ViViT) architecture, incorporating early, intermediate, and late fusion strategies. Initial analysis was conducted on the mannequin data and the first neonatal subject. The results show that for estimating RR in neonates, the early fusion approach outperformed the unimodal methods. In movement detection, compared to human labeling, the fusion techniques achieved high accuracy and precision. To conclude, this study demonstrates that multimodal analysis has the potential to outperform unimodal approaches by improving accuracy against gold standard monitoring, particularly in challenging real-life conditions, including motion artifacts and poor lighting. This work represents a step toward more robust, non-invasive monitoring solutions for neonatal care, with implications for broader applications in remote health monitoring.

Incubators for Species which Exhibit Temperature-Dependent Sex Determination: Application to Hawksbill Sea Turtles in Rising Ambient Temperatures

Thu, 01 May 2025 00:00:00 GMT

Incubators for Species which Exhibit Temperature-Dependent Sex Determination: Application to Hawksbill Sea Turtles in Rising Ambient Temperatures Finlason, Katana R. As global ambient temperatures continue to rise, with the highest recorded annual averages since 1850 being within the last ten years, problems emerge for species exhibiting temperature-dependent sex determination. This is the process by which the sex of an animal’s embryo is determined based on the temperature of environment in which it is incubated, which can result in skewed sex ratios within a population like in the case of the critically endangered Hawksbill Sea Turtles (Eretmochelys imbricata). Reportedly, 85-95% of Hawksbills sampled in the wild are currently female [3]. This sex-imbalance can negatively impact the species’ ability to procreate, leading to the potential for extinction. Currently, no viable, long-term solutions exist to effectively and safely cool sea turtle eggs while still keeping them within their natural habitat. This research proposes the creation of sea turtle egg incubators designed to achieve a temperature range that will produce a higher percentage of male hatchlings to help rectify this imbalance in habitats heavily affected by climate change. These incubators are designed to be affordable, easy to build and, most importantly, safe for the sea turtle eggs. Three-month-long temperature trials for the incubator were conducted in Jamaica with conservationist community partners at Oracabessa Bay Sea Turtle Project. Results showed that this incubator is not only easy to manufacture and use, but that it successfully regulates the temperature range in favor of more male hatchlings, while also increasing the emergence rate of the hatchlings from 70% in natural nests to over 80%. During one of the hottest months in Jamaica, the incubator, deployed without water changes, doubled the predicted percentage of males produced by natural nests. When provided with cool water changes the incubator quintupled this value. Throughout the months of August to October, the incubator achieved a temperature range that is predicted to produce 85-99% male hatchlings, thus counteracting the feminization phenomenon occurring in nature.

Minimal Constraint and Precision Placement in Cyclic Testing of High Life-Cycle Products

Thu, 01 May 2025 00:00:00 GMT

Minimal Constraint and Precision Placement in Cyclic Testing of High Life-Cycle Products Edington, David J. In the electrification of heavy industry, rapid swappable batteries provide an effective means to minimize vehicle downtime and the cost of operation. However, to allow this technology to take hold, further development of electrical contacts that can both pass high amperage and undergo a high cycle life needs to occur. The development of these electrical contacts is a highly experimental process, and thus establishing a method and test equipment to determine the physical and electrical characteristics of these contacts over their lifetime will allow for the accelerated development of these products. This body of work serves as a design guide to establish a physical testing mechanism to assess contact resistance degradation and physical wear over the lifespan of an electric connector. Data will then be collected on initial contact prototypes to characterize their performance. With this data, designs may be iterated and improved upon in pursuit of creating a universal standard for battery swap technology on electric vehicles in heavy industry.

Development of a Hierarchical Reflexive Control Framework for Autonomous Robotic Manipulation

Thu, 01 May 2025 00:00:00 GMT

Development of a Hierarchical Reflexive Control Framework for Autonomous Robotic Manipulation SaLoutos, Andrew Within the field of robotic manipulation, much research focus has been placed on improving perception and planning algorithms, assuming that the actions output by these high-level planners will be easily achieved by the robot systems. However, to surpass human manipulation performance, fast and robust execution of manipulation plans is just as critical as improved perception and planning methods. In this thesis, we introduce the last centimeter problem, which states that the most difficult part of grasp execution is when less than a centimeter remains between fingertips and an object, and contact is imminent. To solve this problem, we propose a reflexive control framework, which is a manipulation control architecture that decouples low-level, high-bandwidth behaviors, which we call reflexes, from broad high-level plans. The reflexes are fast, autonomous reactions to local sensing information that are designed to add robustness to high-level manipulation plans while also reducing the necessary complexity of manipulation planning problems. To deploy our reflexes, we design hardware platforms that incorporate high-bandwidth actuation and low-latency tactile sensing, allowing us to maximize the reactive capabilities of the overall manipulation system. We validate our approach through studies on teleoperated grasping and autonomous planar grasping, which show that our reflexive controllers increase manipulation speed and robustness. Then, we perform extensive simulation studies for autonomous grasping in SE(3), conducting experiments with single objects as well as cluttered scenes, using a variety of state-of-the-art grasp planners. Our results show greatly improved grasp robustness with our reflexive controllers, across all object types and grasp planners. Further experiments show that the benefits of our reflexes persist across sets of objects that are larger, heavier, and more slippery, and with increasing magnitudes of errors in the executed grasp poses. While this thesis demonstrates that the reflexive control framework is effective at increasing grasp robustness during picking, our framework is constructed in a way that is amenable to extension to other tasks, like in-hand manipulation or constrained object placement, as well as application to more complex grippers, such as those with three or more dexterous fingers and more diverse sensing.

Distinct roles for energy storage and transmission infrastructure in a renewables-based electric power system

Thu, 01 May 2025 00:00:00 GMT

Distinct roles for energy storage and transmission infrastructure in a renewables-based electric power system Kim, Beomjun Due to the intermittency of renewable resources, achieving a high coverage of renewable generation at low cost is one of the main hurdles to realizing zero-carbon electricity generation. In this study, we analyze the roles of energy storage systems (ESS) and transmission infrastructure in the cost-optimal deployment of a renewable electricity grid in the United States. We find that storage and transmission serve distinctly different functions: transmission is useful for addressing hours-long resource lows, but only plays a supplementary role in mitigating long-duration resource lows. Conversely, storage can handle both short-duration and long-duration resource lows. These different functions are driven in part by the large spatial footprints of the most extreme long duration resource lows. Furthermore, the total cost of renewable energy in the system and the cost-determining technological components in the system are dependent on the renewables penetration toward total demand—known as the energy availability factor (EAF). When the EAF is sufficiently low, the cost of a cost-optimized system is driven solely by generation costs. For low to intermediate EAF, both generation and transmission costs are dominant factors. At high EAF, generation and storage costs become the dominant factors.

Wedged Vortex Generator Applications for Marine Vessels

Thu, 01 May 2025 00:00:00 GMT

Wedged Vortex Generator Applications for Marine Vessels Kimmeth, Jack This thesis investigates the effectiveness of vortex generators (VGs) in reducing viscous drag in hydrodynamic applications. Initial experimental and computational fluid dynamics analyses identified wedge-shaped VGs as the optimal design for flow manipulation. Comparative testing of three wedge shaped VG sizes at 1.3 m/s revealed the most effective configuration, which was subsequently evaluated across speeds ranging from 1.0 m/s to 1.6 m/s. The results showed a viscous drag reduction of 7.9% at 1.4 m/s. These findings were extrapolated to a full-scale bulk carrier using appropriate geometric and dynamic scaling factors. Total resistance was partitioned using Holtrop-Mennen approximations, allowing the drag reduction to be realistically applied to operational conditions on a trans-Pacific route. Material and installation cost estimates were also developed. Finally, implications for propulsion efficiency, flow-induced vibrations, and cavitation are discussed, with recommendations for future self-propelled model testing to further explore these effects.

Prosody in Kichwa

Thu, 01 May 2025 00:00:00 GMT

Prosody in Kichwa Chango Masaquiza, Soledad This thesis investigates the prosodic system of Salasaka Kichwa, focusing on the interaction between pitch, morphosyntactic structure, and word order in both elicited and spontaneous speech. Based on data from ten native speakers of the Salasaka community, the study analyzes approximately 150 utterances using Praat and ToBI-style prosodic annotation. The findings reveal a consistent alignment between the nuclear pitch accent and the leftmost constituent of the verb phrase in neutral declarative sentences, supporting the hypothesis that Salasaka Kichwa exhibits a head-final syntactic structure. This default prosodic alignment is disrupted by the presence of focus-sensitive or interrogative morphemes such as -mi and -chu, which reliably attract the pitch peak regardless of their position in the clause. In ditransitive constructions, pitch prominence consistently targets the dative-marked argument. Accusative-marked objects also receive prominence, but only when modified; in such cases, it is typically the modifying adjective or contrastive element that bears the highest pitch. Overall, the study demonstrates that prosodic prominence in Salasaka Kichwa is not governed by syntactic structure alone. Instead, it emerges from a layered interaction between morphology, information structure, and pragmatic marking offering new insights into how prosody encodes grammatical and communicative functions in underdescribed head-final languages.

Model Predictive Control Approaches for Dynamic Table Tennis Swinging

Thu, 01 May 2025 00:00:00 GMT

Model Predictive Control Approaches for Dynamic Table Tennis Swinging Nguyen, David H. This thesis presents three model predictive control (MPC) formulations for robotic table tennis swinging, addressing the challenge of generating precise, real-time paddle trajectories for dynamic ball interactions. We explore key differences in optimization structure, solver strategy, and real-time implementation, evaluating each approach through hardware experiments that measure strike condition tracking and hit success. The final controller integrates the full task of a table tennis possession by planning the return ball trajectory through the contact dynamics, and generating a swing to achieve it. This controller improves the hit rate of the system from 88.3% to 97.6% and significantly enhances strike condition accuracy and smoothness enabling control over the landing location and spin of the ball.

On the Certification of Deep Learning-based Dynamical System Identification

Thu, 01 May 2025 00:00:00 GMT

On the Certification of Deep Learning-based Dynamical System Identification Zhang, Wang Dynamical system identification, the reconstruction of the system governing equations from observations, has been studied for decades. With the recent emergence of deep learning techniques, neural network-based parameterization enriches this classical field by offering new capabilities in modeling complex systems. While promising advances have been made, these black box models face significant challenges due to their limited interpretability and lack of physical guarantees, raising concerns about their applicability in scenarios where trustworthiness is critical. In this thesis, we developed a comprehensive framework to analyze, understand and learn dynamical systems. We start with a contrastive learning method to capture system invariants (i.e., conserved quantities) from trajectory observation of dynamical systems. Building on these learned invariants or known priors, we introduce a projection layer for neural networks that guarantees the preservation of physics constraints in the learned dynamics models. This two-step approach significantly improves the trustworthiness and interpretability of the traditional black-box models. On top of this, we extend this methodology to learn physically meaningful embeddings corresponding to inter-system characteristics, enabling zero-shot meta-learning capabilities for dynamical system models. Finally, we reduce the bias gap in the classical neural network-based aleatoric uncertainty estimators. We identify overestimation issues in existing variance attenuation methods and propose a novel denoising-based approach that provides more accurate estimates of data uncertainty. This method not only applies to regression tasks but also extends to dynamical system observations.

Modeling the Impact of Helicopter Vibrations on the Musculoskeletal Health of US Army Blackhawk Helicopter Pilots

Thu, 01 May 2025 00:00:00 GMT

Modeling the Impact of Helicopter Vibrations on the Musculoskeletal Health of US Army Blackhawk Helicopter Pilots Johnston, Julie E. The UH-60, used for troop transport, MEDVAC, and mission control, has evolved over the last 45 years from the Alpha Model to the Lima and Mike models that are currently utilized. Previous studies investigated the impact of Whole-Body Vibrations (WBV) on aviators and the resulting musculoskeletal injury, but none have investigated the efficacy of the Mike model’s Active Vibration Control System (AVCS) on reducing the impact of helicopter vibrations on musculoskeletal health. Computational analyses of a biomechanical model using OpenSim and motion capture at varying levels of vibration was conducted. This quantifies the response of the spine and the surrounding muscles when vibratory loads are applied while positioned to manipulate the flight controls. A musculoskeletal model was developed to represent the aviator in the seated posture required to effectively manipulate the flight controls. To develop the model, the team recorded motion capture data with a pilot in a pilot test for concept validation. This data was then processed and input in the OpenSim inverse kinematics tool to determine joint angle and to demonstrate the muscle-tendon length of several muscles in the back. Unlike the initial predictions, the muscles in the right side of the back were not consistently longer than those of the left side. A survey was also developed that builds upon previous efforts, seeking to understand the aviator’s perspective on musculoskeletal injury and prevention, with a focus on the back. Aviators are asked to describe the cause of their injury, methods of injury prevention, and recovery techniques encompassing numerous subpopulations of flight experience: Lima-majority, Mike-only, Mike-majority, and an even mixture of L/M. The data attempts to characterize the impact of the AVCS on aviator spine health. The AVCS should decrease the rate of injury by reducing the vibratory loads experienced by the aviator. This survey is unique to previous questionnaires as it focuses on the user’s perspective of differences between the two models, and the injury or pain felt by each service member. While it was expected to see a trend of reduced injury occurrence amongst the Mike-only aviators versus those with Lima-majority flight hours, this was not the case. Injury prevalence was consistent across most populations, indicating the potential inefficacy of the AVCS. Analysis of open-ended responses, particularly from the hybrid group, provide some context for the perceived impacts of using the AVCS. Some population demographics were not represented in this survey due to the nature of the unit being surveyed, which may impact the validity of some results. By quantifying the perceived efficacy of the AVCS as it relates to chronic musculoskeletal injury using a survey of pilot experience factors (flight hours, airframes, operating theatres, etc.), and by representing the maladaptive posture of the pilots with a computational simulation based on experimental pilot data; a full picture is developed of the risk of issue related to the near and long-term health of US Army Aviators. The aim is to expand the overall understanding of how vibration is impacting the musculoskeletal health of aviators and their perceived impact on lifelong health from the profession. The ultimate goal is to aid in the design of additional countermeasures to improve aviator spine health and to serve as a platform for optimization of systems like AVCS.

Design Theories for Compact, Low-energy, Clog-resistant Drip Irrigation Emitters

Thu, 01 May 2025 00:00:00 GMT

Design Theories for Compact, Low-energy, Clog-resistant Drip Irrigation Emitters Ghodgaonkar, Aditya This thesis presents the derivation, experimental validation, and demonstration of new design theories for compact, low-pressure, clog-resistant drip emitters that can make drip irrigation affordable, reliable, and easier for farmers to adopt. Broad adoption of water-efficient irrigation methods such as drip irrigation is imperative to sustainably meet projected global food demand against the backdrop of diminishing freshwater resources, constrained arable land, and climate change. In drip irrigation systems, emitters are passive flow-regulating devices that are inserted into the drip tube to align with every plant. They are designed to provide a constant flow rate once they are pressurized to at least their activation pressure, thus ensuring uniform, localized irrigation of plants. However, conventional emitters directly contribute to three barriers that have limited drip irrigation adoption – high raw material-driven equipment costs, high pumping power costs associated with pressurizing all emitters in the field to their activation pressure, and gradual loss of reliability due to clogging. Compact, low-pressure, clog-resistant emitters can address these challenges, but to design them, we must model and tune their operating physics, which is centered around two complex features – a millimeter-scale tortuous passage called the labyrinth, and fluid-structure interaction (FSI) involving a flexible silicone rubber diaphragm and a micro-duct. This makes conventional design approaches relying on high-fidelity simulation software or empirical trial-and-error too expensive and time-consuming to use for the development of compact, low-pressure, clog-resistant emitters on competitive industrial timelines. This thesis addresses these challenges through three contributions. The first contribution presents an empirically derived hydraulic model of emitter labyrinths, which are typically the most volume-intensive feature of emitters. The model relates labyrinth flow rate to select material volume agnostic parameters, allowing designers to create compact labyrinths with desired hydraulic performance. The compact labyrinths can enable up to 10% reduction in the raw material-driven cost of drip equipment. The second contribution presents a 1-dimensional model of the FSI in emitters that can predict their flow rate-pressure performance in 2-3 minutes and within 8-14% error, cutting down on design cycle times by orders of magnitude. This facilitated the rapid synthesis of low-pressure emitter designs having 50-60% less activation pressure than conventional emitters, cutting pumping power costs by an estimated 18-23%. Together, the first two contributions can enable an estimated 18% reduction in the lifetime costs of drip irrigation, but long-term adoption requires that the emitters be clog-resistant and compatible with the current maintenance practices of farmers. To that end, the third contribution presents an experimental investigation of clogging in low-pressure emitters. The results of the investigation directly correlated the geometry of emitter hydraulic features to the critical particle size that would clog them. As a result, compact, low-pressure emitters could be designed to be compatible with the same filters and maintenance practices as current state-of-the-art products that have higher activation pressures. This was confirmed by field testing the compact, low-pressure, clog-resistant (MIT) emitters alongside commercial reference designs with their prescribed filters for nearly 1200 hours. At the end of the field test, the MIT emitters still held 90-94% of their initial flow rate, putting them on par with or better than the reference products in terms of irrigation reliability. The collective contributions of this thesis present the knowledge needed to design emitters that can make drip irrigation more affordable to adopt by farmers and demonstrate that substantial capital and operating cost reductions can be realized without sacrificing product reliability or requiring expensive changes to current farmer maintenance practices.

Hydrodynamic Behavior of Pop-Up Satellite Archival Tags (PSAT) Subject to Vortices

Thu, 01 May 2025 00:00:00 GMT

Hydrodynamic Behavior of Pop-Up Satellite Archival Tags (PSAT) Subject to Vortices Hoo, Stephanie Pop-up Satellite Archival Tags (PSATs) are a combination of satellite and archival tags used by marine biologists to collect large scale movement and behavioral data of large pelagic life for up to two years [1]. However, current commercial PSATs have an unusually high failure rate when tagged on tuna and cost upwards of $4000, making it both difficult and expensive to collect data [14]. Upon investigation, the top two failure modes of tuna-affixed PSATs have been identified as drag from movement/tissue healing and pressure cycling [14]. Current commercial PSAT manufacturers do not account for the vortices shed by fish when testing their designs— a large oversight that could account for their high failure rate [15]. The work herein determined the effects of vortex shedding on PSAT hydrodynamic behavior, used these results to inform the design of novel PSAT body shapes, and conducted a head-to-head comparison of these designs with existing commercial PSATs.

Combating Corrosion and Monitoring Microgrids on Coast Guard Patrol Boats

Thu, 01 May 2025 00:00:00 GMT

Combating Corrosion and Monitoring Microgrids on Coast Guard Patrol Boats Buchanan, Maxwell Calvin Marine corrosion presents a persistent threat to the reliable operation of U.S. Coast Guard Fast Response Cutters (FRCs). This thesis investigates hybrid cathodic protection strategies combining impressed current cathodic protection (ICCP) systems and sacrificial zinc anodes to combat corrosion on such vessels. Observing over 550 cumulative months of ICCP system data across 46 FRCs, this thesis identifies operational trends, failure modes, and unique regional behaviors. To validate observed patterns and explore failure scenarios, the study implements finite element modeling using COMSOL Multiphysics. These simulations replicate normal operation, reference electrode failure, propeller passivation, localized zinc loss, and hull coating failure for both a generic 35m hull and the FRC hull. These models emphasize how system behavior responds to material variations, temperature, and system health, offering a diagnostic framework for optimizing ICCP configurations. Field and laboratory experiments further ground the computational findings. These include shipboard hull potential surveys and analysis of zinc anode wastage across multiple cutters. Controlled experiments on nickel aluminum bronze (NAB) passivation using miniaturized ICCP test systems are explored for further study. Initial results show variation in zinc consumption and corrosion behavior depending on ICCP setpoints, with higher protection levels (-1050 mV) often correlating with reduced zinc depletion. The thesis also explores energy diagnostics onboard FRCs via non-intrusive load monitoring (NILM). A case study on the USCGC WILLIAM CHADWICK describes monitoring auxiliary machinery loads through NILM signatures and suggests expansion to critical panels and DC systems. By integrating fleet data, physical experimentation, and simulation, this thesis advances future efforts in patrol boat corrosion monitoring, ICCP optimization, and resilient microgrid management.

Incorporating High-Resolution Tactile Perception for Performative and Generalized Robotic Manipulation Through Compliance Estimation and Hardware Design

Thu, 01 May 2025 00:00:00 GMT

Incorporating High-Resolution Tactile Perception for Performative and Generalized Robotic Manipulation Through Compliance Estimation and Hardware Design Burgess, Michael In robotics, replicating the natural proficiency with which humans perform manipulation tasks has proven challenging. Modern control schemes are predominantly learning-based and thus depend heavily on data collected via teleoperated demonstrations. Humans rely on our tactile perception to perform contact-rich and dynamic manipulation tasks. By more seamlessly incorporating high-resolution tactile sensing and haptic feedback into teleoperation interfaces, we can work to create stronger demonstration data to support the development of more effective learned control policies. In this thesis, we present two contributions toward this goal. First, we develop an algorithm to estimate the compliance of grasped objects in real-time from tactile images to provide haptic feedback to remote users. This algorithm combines both analytical and learning-based approaches to better generalize across both object shapes and materials. Second, we create a 1-DoF robotic gripper design with integrated tactile sensing. Inspired by the principle of self-similarity, this gripper is designed to better conform to complex object geometries than traditional designs and more securely grasp objects of many shapes and sizes. Together, these contributions can be utilized to create robust, tactile-aware teleoperation platforms. These platforms would facilitate more effective data collection and thereby promote the development of more performative autonomous action in generalized robotic manipulation scenarios.

Tailoring Complexity of Model-Based Controllers for Legged Robots

Thu, 01 May 2025 00:00:00 GMT

Tailoring Complexity of Model-Based Controllers for Legged Robots Khazoom, Charles Humanoid robots promise human-like mobility, but must manage complex and often conflicting control objectives. While model-based controllers can address these challenges using online optimization, they have high computational demands. Model predictive control (MPC) provides closed-loop stability with online trajectory optimization, but achieving real-time rates is difficult for high-dimensional systems. To mitigate this limitation, most MPC implementations rely on reduced-order models (ROMs) that simplify planning but fail to capture whole-body constraints like joint limits and self-collisions. Reactive whole-body controllers (WBCs) partially address this limitation by projecting ROM trajectories onto some wholebody constraints, but these are restricted to acceleration-level constraints like friction cones and torque limits. This thesis advances humanoid planning and control through a renewed focus on model fidelity, solution accuracy ans solve times with three key contributions. First, we propose the CBF-WBC, which augments reactive WBCs with position constraints using control barrier functions (CBFs), enabling the MIT Humanoid to avoid selfcollisions with minimal computational overhead. As a result, the robot can reactively deviate from infeasible trajectories from a reduced-order MPC. Despite fast solve times below 100 microseconds, conflicts can arise between the reduced-order MPC and the CBF-WBC. To address this, we enable real-time whole-body MPC using the alternating direction method of multipliers (ADMM) to provide low-accuracy solutions at high feedback rates. The controller is reliably deployed on hardware and enables the MIT Humanoid to walk robustly on rough terrains and plan complex crossed-leg and arm motions that enhance stability when recovering from significant disturbances. While low-accuracy solutions often suffice for real-time control, we found that higher accuracy could still improve closed-loop performance if computational speed allows. Building on this insight, we propose a framework to simultaneously optimize solution accuracy and model complexity to maximize closed-loop performance. Instead of planning with a single model that is too complex or too simple, solve times can be reduced by planning over a sequence of models of reducing complexity. We extract ROMs from whole-body dynamics equations and optimize their horizons, discretization timesteps and solution accuracy using blackbox optimization. The optimizer can sacrifice model complexity for additional ADMM iterations, reducing falls by nine-fold and enabling a 2 m/s walking speed on hardware.

The Net Climate Impact of AI: Balancing Current Costs with Future Climate Benefits

Thu, 01 May 2025 00:00:00 GMT

The Net Climate Impact of AI: Balancing Current Costs with Future Climate Benefits Turliuk, Jennifer What is the net impact of artificial intelligence on climate change? Existing studies focus on AI's footprint, but few analyze AI's trade-offs. This paper develops a framework to quantify both the Greenhouse Gas (GHG) emissions and the climate change costs and benefits of AI systems, addressing the time value of carbon and the installed base of existing AI infrastructure. We examine the energy demands of AI, which are growing rapidly and threatening companies' net-zero commitments, while also analyzing AI's potential to enable emissions reductions through applications such as optimized energy systems, demand response, grid management, and electrification acceleration. This research introduces the Net Climate Impact Score (NCIS) of AI, a novel equation to calculate the net climate impact of AI technologies that considers both immediate emissions and potential future benefits, and provides a methodology for assessing AI projects holistically. We demonstrate that while current AI applications are predominantly emissions-intensive, strategic deployment focused on energy system transformation could potentially deliver net climate benefits within specific time frames and applications. However, improvements in energy efficiency and emissions reductions resulting from AI are, absent climate policy, likely to generate both direct and indirect rebound effects that could undermine the emissions reductions and reduce the climate benefits of AI. The research concludes with policy and industry recommendations that propose technological pathways that could maximize AI's positive impact while minimizing its environmental footprint.

Wide Range Switched Mode RF Power Amplifiers and their applications

Thu, 01 May 2025 00:00:00 GMT

Wide Range Switched Mode RF Power Amplifiers and their applications Pressel, Adam Jay Switched-mode power amplifiers (SMPAs) are desired that can work across a wide range of power levels and load impedances with fast response speed while maintaining high efficiency. Such designs would be valuable for many applications including plasma generation and wireless power transfer. We introduce a new wide-range SMPA architecture that provides direct output voltage modulation, enabling it to modulate output power and compensate for resistive load variations. Dynamic frequency modulation is leveraged to address reactive load variations. The new architecture enables all the semiconductor switches to maintain zero-voltage switching across all operating conditions. Experimental results shows that the wide-range half bridge power amplifier was able to deliver a wide power range of 25 W - 95 W power range across each individual resistive load in the range of 5 Ω - 20 Ω with up to j15 Ω reactance. The maximum dc-ac efficiency is 86 with 20 Ω load and 110.5 W load power.

Tension-Leg Platform for Offshore Diffusor-AugmentedHydrokinetic Turbine

Thu, 01 May 2025 00:00:00 GMT

Tension-Leg Platform for Offshore Diffusor-AugmentedHydrokinetic Turbine Mannier, Robert B. Harnessing marine energy offers significant potential for advancing clean and sustainable power generation. This thesis focuses on the design and optimization of a diffuser-augmented hydrokinetic turbine, supported by a tension-leg platform, to harness ocean and tidal currents for renewable energy production. By incorporating diffuser technology, the turbine’s efficiency is enhanced, increasing the coefficient of power and enabling effective energy capture even in environments with lower current speeds. The research involves 2D and 2D axisymmetric modeling of the diffuser and turbine using Actuator Disk Theory (ADT), with tools such as Rhino and Star CCM+. Mounted on a floating tension-leg platform anchored to the seabed, the turbine is designed to exceed the Betz limit, maximizing power output and advancing offshore energy harvesting capabilities. This thesis is solely focused on the design and optimization of the hydrokinetic turbine, providing an in-depth analysis of diffuser performance. The findings contribute to the development of marine renewable energy technologies, promoting sustainable and efficient power generation from ocean and tidal currents.

Optimized Sustainable Hydrogen Generation from Liquid Metal Activated Aluminum-Water Reactions

Thu, 01 May 2025 00:00:00 GMT

Optimized Sustainable Hydrogen Generation from Liquid Metal Activated Aluminum-Water Reactions Kombargi, Aly This study presents a sustainable and cost-effective method for hydrogen generation using aluminum waste, addressing both energy and environmental challenges. Activated aluminum reacts with water to produce hydrogen, heat, and aluminum oxyhydroxide (boehmite), a commercially valuable byproduct. As a safe, efficient, and cost-effective energy carrier with an energy density exceeding 20 kWh/L (8 kWh/kg), aluminum enables on-demand hydrogen production for diverse applications, including maritime transport and off-grid power systems. This research optimizes reaction kinetics to enhance hydrogen yield and rate while minimizing costs and carbon emissions. Activation involves coating aluminum with a gallium-indium eutectic (eGaIn) liquid metal, which disrupts the oxide layer and enables spontaneous reaction in aqueous environments. The study investigates seawater as an ionic medium for eGaIn eutectic agglomeration and reuse. However, chlorine binding slows the reaction, which was countered using high-temperature operation and catalytic enhancement. Adding 0.02 M imidazole accelerated the reaction 60-fold, enabled 92% eutectic recovery, and achieved 99% of the theoretical hydrogen yield. Environmental conditions significantly influence reaction efficiency. Increasing seawater temperature from 20°C to 90°C enhanced reaction rates 44-fold, aligning with Arrhenius Law. Isochoric reactions at high pressure were tested to simulate deep-sea vehicle environments using onboard hydrogen reactors fueled by aluminum and surrounding seawater. Results showed a 33% yield increase at 6 MPa (586 m depth) compared to atmospheric pressure, primarily due to surface tension effects that reduce hydrogen bubble size, improving aluminum-water contact at higher pressures. A life cycle and cost analysis identified an optimized production scenario with a carbon footprint of 1.45 kgCO2eq/kg H2, meeting green hydrogen standards. Major contributors include recycled aluminum use and processing, and the eGaIn alloy; but eutectic recovery and thermal energy reuse further reduce emissions. Using scrap aluminum and recovering byproducts, hydrogen production costs are estimated at $9.2/kg. Additionally, reselling boehmite (market price $2.5/kg) could generate revenue 5.6 times greater than input costs, significantly improving economic viability. To demonstrate scalability, a modular hydrogen reactor was developed and directly integrated with a commercial generator, reliably producing 400W of power from on-demand, 99% purity lab-tested hydrogen. The envisioned application is a fully integrated aluminum recycling system that utilizes aluminum waste and seawater to generate hydrogen, thermal energy, and boehmite. This approach advances clean energy technology by providing a scalable and economically viable hydrogen production pathway. Beyond its direct application in underwater technologies, this optimized reaction can support energy-intensive operations such as heating, desalination, transportation, industrial hydrogen production for refining and fertilizer synthesis, stationary energy systems for off-grid power, and renewable energy storage. Its versatility strengthens energy security and decarbonization efforts while offering a cost-competitive alternative to conventional fuels, positioning it as a key enabler of a sustainable energy future.

Essays in Venture Capital and Corporate Finance

Thu, 01 May 2025 00:00:00 GMT

Essays in Venture Capital and Corporate Finance Paine, Fiona This thesis is three chapters. In the first chapter, I study the impact of restricting foreign venture capital investments for national security reasons. Countries have increasingly been using economic policies to further geopolitical and national security goals. Thus far, economists have focused on studying tariffs and subsidies despite a broader range of economic tools actually being implemented. How costly are these other policies and what are their effects on capital markets, investment, and the economy more broadly? In this paper, I examine a 2018 U.S. law (FIRRMA), which expanded the government’s ability to review and block transactions on national security grounds to include venture capital (VC) investments by foreign investors. I use the passing of FIRRMA, its differential impact on specific VC industries, and the role of Chinese investors in U.S. venture capital to study whether foreign investment screening impacts capital supply. I find that FIRRMA had a negative effect on capital supply in impacted industries due to two factors: 1) the specialization of VC investing (such that the substitution of outside capital into impacted industries is low) and 2) networks in VC investing (there are spillovers to domestic syndication partners in impacted industries). I further find that the change in capital supply is costly, leading to lower innovation by startups. I introduce a novel way of measuring innovation early in the life of a startup using text from startup websites. I use this measure to show there is a selection effect where VCs give first round funding to less innovative startups after FIRRMA. Finally, in a case study of the biotechnology industry, I show that impacted startups suspend drug projects at higher rates, and in particular their risky projects. In the second chapter, joint with Johnathan Jensen, we study municipal cyber risk. Cyber attacks are estimated to cost billions of dollars per year. However, cyber risk is hard to study since companies rarely disclose hacks and don’t share information on cyber security investment. This paper takes a novel approach by looking at municipal hacking. We use a dataset of municipal ransomware attacks merged with hand collected IT investment data and municipal bond data. We find that lower IT investment predicts hacking. Furthermore, following a ransomware attack, municipal bond yields fall by 13 basis points and IT investment as a share of total town expenditure increases by 23 basis points. We investigate potential channels leading to decreased yields post hacking. We find evidence that being hacked reduces cyber risk by disciplining municipalities to move closer to the optimal level of IT spending. The third chapter investigates the impact of firm data collection and analysis of collected data on the riskiness of firm cash flows. I use a scraped data set of the third party resources loaded on firms’ websites as a measure of firm data collection and analysis practices. I find that firm u se of less effective web analytics is as sociated with an increase in the variance of sales, inventory, and both fixed and variable costs. This effect is de spite a lack of change in the level of these variables. Looking at the effect of treatment on the treated, there i s higher profit and sales variance during times of higher uncertainty. I use differences in web analytics technology and a change in their relative effectiveness as my identification strategy. As a case study of a large negative demand shock, I look at differences in fi rm reactions to COVID-19 based on their web analytics usage.

Coevolution of Small Business Strategy and Regulation: A Mixed-Methods Study of United States Craft Breweries

Thu, 01 May 2025 00:00:00 GMT

Coevolution of Small Business Strategy and Regulation: A Mixed-Methods Study of United States Craft Breweries Rixey V, Eppa This dissertation asks how do small firms overcome regulatory constraints despite powerful opposition? Significant research has documented the nonmarket strategies of large, multinational firms seeking to benefit from and capture regulatory systems. However, despite the historically important role of small and medium-sized enterprises (SMEs) in the economic and civic structures of the US, there is much we do not know about whether and how they attempt to exert their own influence in regulatory environments. To explore this, the US beer industry was selected as a strategic research site where SMEs have had a range of successes and failures in developing policy influence. In the late 1970s, the US beer industry rapidly consolidated to less than 100 breweries, but today, with the rise of small, craft breweries, there are over 9,000 breweries in the US. Over 7,000 of these focus on direct-to-consumer (DTC) sales, which were explicitly or practically illegal in all 50 states in 1980. How did this market and regulatory transformation take place and why did some states significantly change their policies to support small brewers while others did not? Two studies were conducted to explore this, an in-depth qualitative study of a single state and a mixed-methods comparative study of six states. The single state was selected for variation in policy outcomes over time and at local levels. Through interviews and archival research, it was revealed that craft breweries engaged in a bottoms-up approach, through which individual firms venue shift downwards, from state to local regulators, to successfully ease state-level constraints. In local public hearings, individual entrepreneurs blended local corporate social responsibility (CSR) with an experimental approach to corporate political activity (CPA) that motivated city-based regulators to challenge state-level restrictions on DTC business models. To understand how this process of developing policy influence unfolds in the absence of local regulators, the national trade associations in the beer industry were analyzed and six states where the state has near exclusive control over alcohol regulations were selected for further analysis. Controlling for a range of factors through a cross-sectional database led to a geographically proximate sample of six comparable states with wide variation in the favorability of policies and the number of breweries per capita. A unique dataset of over 5,000 legislative updates on proposed and enacted federal and state policy changes was supplemented with archival and interview data to assess policy influence. The conventional approach described in the literature, collective action via a trade association, was important but often insufficient. Each state had a functioning trade association representing most craft breweries, but sustained policy influence was observed only in states where full-time leaders of these associations understood the political landscape and developed policy partnerships to tilt the odds in their favor. Policy partnerships entailed legislation alleviating regulatory constraints while also including new provisions that ensured long-term alignment among the partners. Taken together, these studies reveal the vital importance of collective action extending beyond the focal industry for SMEs to develop policy influence at the local or state level.

On the Application of an Output-based Adaptive, Higher-order Finite Element Method to Sonic Boom Propagation

Sun, 01 Sep 2024 00:00:00 GMT

On the Application of an Output-based Adaptive, Higher-order Finite Element Method to Sonic Boom Propagation Trono Figueras, Renato The reduction of sonic boom loudness to within acceptable limits is a crucial factor for the viability of supersonic aircraft. This thesis presents a computational framework for simulating sonic boom propagation using an output-based adaptive, higher-order finite element method. The research employs the Variational Multiscale with Discontinuous Subscales (VMSD) method, integrating Continuous Galerkin (CG) and Discontinuous Galerkin (DG) features, referred to as VMSD-BR2. This approach leverages static condensation to manage computational cost while utilizing DG stabilization techniques for enhanced stability and adjoint consistency. A key component of this work is the application of the dual weighted residual (DWR) method for output error estimation, which in turns drives the mesh optimization process. The method’s efficacy is validated using smooth solutions for the viscous Burgers equation and the adjoint PDE for a volume output functional. Additionally, artificial viscosity is incorporated via a shock sensor PDE approach to handle shock presence, with necessary corrections applied to the DWR error estimate. The VMSD-BR2 method is applied then to a real-world scenario solving the augmented Burgers equation, which models the propagation of sonic booms. The results include the pressure perturbation field, adapted meshes, ground-level B-SEL filtered pressure, and perceived loudness at ground, demonstrating the method’s practical application.

C. elegans as a Platform for Multimodal Neural Data Integration

Thu, 01 May 2025 00:00:00 GMT

C. elegans as a Platform for Multimodal Neural Data Integration Simeon, Quilee Systems neuroscience has traditionally been fragmented into investigations at discrete levels of organization, creating methodological and conceptual gaps that hinder unified understanding of neural function. This thesis examines the nematode Caenorhabditis elegans as a platform for integrating diverse neural data modalities, offering a pathway to bridge these gaps. The hermaphrodite C. elegans, with its completely mapped connectome, optical transparency, genetic tractability, and stereotyped nervous system of only 302 neurons, presents an opportunity for comprehensive measurements across multiple dimensions of neural function. The review is organized around three fundamental neural data modalities accessible in C. elegans: (1) molecular genetic profiles, (2) network connectivity, and (3) neural activity dynamics. Historically studied in isolation, these complementary data types are increasingly being bridged through technological and computational innovations. We examine experimental advances enabling whole-nervous-system measurements of these modalities, as well as data standardization efforts and computational frameworks for cross-modal integration. While understanding the relationship between neural activity and behavior remains a fundamental goal of systems neuroscience, this thesis focuses on neural data acquisition and integration rather than behavioral analysis, which has been extensively covered elsewhere.1 We conclude with some original proposals to overcome current limitations in multimodal data acquisition and synthesis, and suggest future directions toward a holistic understanding of how molecular components, network connectivity, and cellular physiology collectively give rise to neural function in C. elegans. These integrative approaches establish a roadmap that may eventually scale to more complex nervous systems and advance our understanding of neural computation across species.

A Nickel Short: Rethinking Element Scarcity in Pursuit of a Fusion-Powered World

Thu, 01 May 2025 00:00:00 GMT

A Nickel Short: Rethinking Element Scarcity in Pursuit of a Fusion-Powered World Sutcliffe, Douglas Fusion energy presents a promising solution for current global decarbonization goals. This thesis presents an adaptable model for evaluating mineral sufficiency in the global deployment of fusion power. Using the ARC Magnetic Confinement (MC) Deuterium-Tritium (D-T) fusion concept as a framework, this research integrates mineral usage estimates from the International Energy Agency (IEA) with MIT Energy Initiative’s (MITEI) energy production forecasts by generation technology. Using MITEI’s $2,800/kW cost scenario for fusion power generation, the model situates the demand for fusion-critical minerals within the broader context of growing mineral needs driven by the clean energy transition, and offers specific, quantitative insights into mineral sufficiency risks. The study finds that beryllium will face significant shortages solely due to fusion demand, with resource exhaustion projected to occur within 40 years. When accounting for additional demands from Electric Vehicles (EVs), battery storage, and transmission infrastructure, chromium and nickel are projected to exhaust economically extractable reserves within 21 to 35 years at current prices. The research further reveals that for nine of the thirty elements evaluated, over 50% of production is concentrated in a single country, and for half of the minerals China is the largest producer, introducing geopolitical risks. Notably, at just 13 kg per reactor, the demand for Rare Earth Elements (REEs) is not exposed to a significant risk, even without the top producing country. The research also surfaces current reactor designs and strategies which could help mitigate each identified risk.

As global ambient temperatures continue to rise, with the highest recorded annual averages since 1850 being within the last ten years, problems emerge for species exhibiting temperature-dependent sex determination. This is the process by which the sex of an animal’s embryo is determined based on the temperature of environment in which it is incubated, which can result in skewed sex ratios within a population like in the case of the critically endangered Hawksbill Sea Turtles (Eretmochelys imbricata). Reportedly, 85-95% of Hawksbills sampled in the wild are currently female [3]. This sex-imbalance can negatively impact the species’ ability to procreate, leading to the potential for extinction. Currently, no viable, long-term solutions exist to effectively and safely cool sea turtle eggs while still keeping them within their natural habitat. This research proposes the creation of sea turtle egg incubators designed to achieve a temperature range that will produce a higher percentage of male hatchlings to help rectify this imbalance in habitats heavily affected by climate change. These incubators are designed to be affordable, easy to build and, most importantly, safe for the sea turtle eggs. Three-month-long temperature trials for the incubator were conducted in Jamaica with conservationist community partners at Oracabessa Bay Sea Turtle Project. Results showed that this incubator is not only easy to manufacture and use, but that it successfully regulates the temperature range in favor of more male hatchlings, while also increasing the emergence rate of the hatchlings from 70% in natural nests to over 80%. During one of the hottest months in Jamaica, the incubator, deployed without water changes, doubled the predicted percentage of males produced by natural nests. When provided with cool water changes the incubator quintupled this value. Throughout the months of August to October, the incubator achieved a temperature range that is predicted to produce 85-99% male hatchlings, thus counteracting the feminization phenomenon occurring in nature.

Thu, 01 May 2025 00:00:00 GMT

As global ambient temperatures continue to rise, with the highest recorded annual averages since 1850 being within the last ten years, problems emerge for species exhibiting temperature-dependent sex determination. This is the process by which the sex of an animal’s embryo is determined based on the temperature of environment in which it is incubated, which can result in skewed sex ratios within a population like in the case of the critically endangered Hawksbill Sea Turtles (Eretmochelys imbricata). Reportedly, 85-95% of Hawksbills sampled in the wild are currently female [3]. This sex-imbalance can negatively impact the species’ ability to procreate, leading to the potential for extinction. Currently, no viable, long-term solutions exist to effectively and safely cool sea turtle eggs while still keeping them within their natural habitat. This research proposes the creation of sea turtle egg incubators designed to achieve a temperature range that will produce a higher percentage of male hatchlings to help rectify this imbalance in habitats heavily affected by climate change. These incubators are designed to be affordable, easy to build and, most importantly, safe for the sea turtle eggs. Three-month-long temperature trials for the incubator were conducted in Jamaica with conservationist community partners at Oracabessa Bay Sea Turtle Project. Results showed that this incubator is not only easy to manufacture and use, but that it successfully regulates the temperature range in favor of more male hatchlings, while also increasing the emergence rate of the hatchlings from 70% in natural nests to over 80%. During one of the hottest months in Jamaica, the incubator, deployed without water changes, doubled the predicted percentage of males produced by natural nests. When provided with cool water changes the incubator quintupled this value. Throughout the months of August to October, the incubator achieved a temperature range that is predicted to produce 85-99% male hatchlings, thus counteracting the feminization phenomenon occurring in nature. Espinal, Michael A. Foams, widely used in packaging, insulation, protective gear, and medical implants, are versatile materials but mechanically inefficient due to their bending-dominated microstructure, leading to an exponential loss of stiffness and strength at low relative densities. Architected materials address this limitation through engineered microstructures that achieve near-linear scaling of properties with relative density. However, truss- and plate-based designs suffer from stress concentrations, while shell-based architectures, though more mechanically efficient, remain highly sensitive to defects and are challenging to fabricate at scale via additive manufacturing. Spinodal architected materials, derived from scalable spinodal decomposition processes, offer a promising alternative with aperiodic, double-curvature microstructures that enhance mechanical efficiency at low relative densities. Nevertheless, their behavior beyond the elastic regime remains largely unexplored. This thesis investigates the nonlinear mechanics of spinodal architected materials by combining a comprehensive experimental dataset with computational modeling. A total of 107 unique morphologies were fabricated and subjected to uniaxial compression along three principal directions, resulting in a dataset of 321 stress-strain curves. Morphologies were generated via simulated spinodal decomposition, allowing controlled variation of anisotropy. Explicit finite element simulations, validated against experimental data, revealed that plastic energy dissipation dominates the large-strain mechanical response. To quantitatively link local morphology to global mechanical behavior, we introduce the Normal Participation Factor (NPF) — a scalar geometric parameter that captures the alignment between surface normals and the loading direction. We demonstrate that the NPF is a material-agnostic proxy for equivalent plastic strain and is linearly correlated with the total energy dissipated during deformation. Combining insights from both experiments and simulations, we establish the NPF as a first-order predictive tool for mechanical behavior under large strains, enabling structure-property predictions without reliance on costly simulations or extensive experimental testing. Altogether, this work lays the foundation for developing finite-strain structure-property relationships in spinodal architected materials, advancing their potential for real-world applications.

Advancing Tendon-Driven Robotic Systems: From Climbing Robots to String Actuators

Thu, 01 May 2025 00:00:00 GMT

Advancing Tendon-Driven Robotic Systems: From Climbing Robots to String Actuators Poon, Ryan Joseph Mar Tendon-driven mechanisms provide a range of benefits for robotic systems, particularly by allowing actuators to be mounted at the base of a manipulator and reducing its inertia. This thesis explores two projects that exploit and advance tendon-driven mechanisms: a wheeled-grasping hybrid climbing robot with modular tendon-driven grasping arms and a hybrid twisted-winching string actuator. Called CLIMR (Cabled Limb Interlocking Modular Robot), the novel climbing robot adapts to columns of varying diameters by adding or removing modular arm links. CLIMR also features capabilities like self-locking (the ability of the robot to stay on the column without power), autonomous grasping, and rotation around the column axis. Mathematical models describe conditions for self-locking, vertical wheeled climbing, and complete grasping of a column. Simulations and experimental results validate the proposed models. The insights from CLIMR are then extended into general design strategies for future developments of similar hybrid climbing robots, focusing on methods to inform design decisions and assess metrics such as adaptability. Ultimately, this work provides a comprehensive framework for designing hybrid climbing robots, highlighting the potential of autonomous solutions for environments where climbing tall structures is critical. Stemming from this climbing robot work is a novel actuator system combining a twisted string actuator (TSA) with a winch mechanism. Relative to traditional hydraulic and pneumatic systems, TSAs are compact but face limitations in stroke length and velocity. This TSA-winch system overcomes these constraints without risking overtwisting by providing both high displacement winching and high force twisting modes. The design features a rotating turret that houses a winch and a worm gear transmission driven by a through-hole drive shaft. Models are developed for the combined displacement and velocity control of this system. Experiments validate the open loop model as well as the closed loop model, which uses a conductive string feedback controller with a gain scheduling and control effort allocation scheme. For specific cases that require large displacement winching followed by high force twisting over several repeatable cycles, an alternate design sacrifices complete string state control and replaces a motor with passive automatic clutches to achieve a seamless transition between modes triggered by the string load. The models of the clutch torque thresholds for this version of the actuator are verified by experiments. Overall, this research contributes to the development of more versatile and efficient actuation systems for tendon-driven robotic applications.

Coordination of distributed energy resources for a reliable, resilient, and affordable decarbonized grid

Thu, 01 May 2025 00:00:00 GMT

Coordination of distributed energy resources for a reliable, resilient, and affordable decarbonized grid Jagadeesan Nair, Vineet Rapid decarbonization of the power grid is essential to meet climate goals by reducing emissions and enabling sustainable electrification of sectors like transport and heating. This requires shifting from centralized fossil-fuel generation to variable renewables like wind and solar. The grid must also adapt to a growing number of small-scale, distributed energy resources (DERs) at the edge, such as rooftop solar, batteries, electric vehicles, and heat pumps. This thesis focuses on modeling, optimizing, and coordinating DERs to enable a flexible, resilient, and affordable grid. First, it proposes a novel hierarchical local electricity market for low and medium-voltage distribution grids. This structure enables DER participation through decentralized and distributed optimization, respecting grid physics while preserving privacy and scalability. The market is applicable to both balanced and unbalanced radial grids using two different convex relaxations and power flow models. Grid services are also priced based on duality theory. Numerical simulations show improved dispatch efficiency, reliability, voltage regulation, and lower retail electricity rates. Second, the thesis applies game theory and mechanism design to extract flexibility from autonomous, strategic DER owners. A repeated Stackelberg game with incomplete information and intertemporal constraints yields equilibrium pricing with closed-form solutions. Third, a distributed decision-making framework is developed to coordinate DERs for grid resilience. It mitigates cyber-physical attacks and outages, ranging from 5 to 40% of peak load, using local flexibility and grid reconfiguration, extensively validated through both software and hardware-in-the-loop simulations. Finally, the thesis addresses DER hosting capacity. New algorithms are developed that co-optimize the siting and sizing of diverse DERs under uncertainty using Monte Carlo sampling, stochastic programming, and k-means clustering for scenario reduction. Results show that intelligent DER coordination can defer grid infrastructure upgrades and support greater renewable integration and electrified demand growth. Together, these contributions provide analytical and simulation tools to improve the planning and real-time operation of future distributed, low-carbon power grids.

The Development and Utilization of Tandem Fluency in Human-Exoskeleton Interaction

Thu, 01 May 2025 00:00:00 GMT

The Development and Utilization of Tandem Fluency in Human-Exoskeleton Interaction Koo, Bon H. (Brandon) There is strong demand for portable technologies that enhance human power output while maintaining safety and range, not only in defense and industry but also in aerospace. Exoskeletons and other wearable powered devices have been proposed as solutions, but a major barrier to adoption is the issue of “fluency”: a combination of metrics representing the seamlessness of human-robot interaction. Most current exoskeleton systems, especially for non-cyclic motions, disrupt user intent and movement, often offering no benefit, or even causing harm by increasing discomfort and injury risk. This lack of fluency is frequently linked to poor intent recognition and absence of predictive control. To address this, we propose developing a human motion prediction system and studying its impact on fluency in exoskeleton-like devices and related human-centered technologies in real-world applications. We introduce an expanded metric “tandem fluency” based on conventional fluency, tailored for evaluating human-robot interaction (HRI) systems where human and robot agents are kinematically synchronized to perform functional tasks. We then develop a proof-of-concept and a functional deep neural network (DNN) capable of detecting human motion intent and predicting motion trajectories in advance using biosignals such as surface electromyography (sEMG). In parallel, we build and test prototype exoskeleton hardware with both single and multiple degrees of freedom. Finally, we conduct human trials with the full closed-loop tandem human-exoskeleton system to evaluate the impact of motion prediction-based control on tandem fluency. The results show that classification and regression prediction of human motion prior to initiation of physical motion is possible and can have performance necessary for practical application of this information, the prediction can be generated not only prior to the physical motion initiation, but often even before the full electrical activation of the primary agonist in many motions, the DNN is robust to variations in sensor hardware and input formatting, and furthermore the use of this prediction in the controls of a tandem robot system has potential to improve tandem fluency by positively affecting both subjective experience and objective/metabolic results.

Mapping Informality: An Approach to Classifying Sidewalk Informal Practices and Elements Through Street View Imagery

Thu, 01 May 2025 00:00:00 GMT

Mapping Informality: An Approach to Classifying Sidewalk Informal Practices and Elements Through Street View Imagery Co, Dominic Lim By 2050, the United Nations estimates that 68 percent of the world’s population will live in cities, with 90 percent of that growth concentrated in rapidly urbanizing informal communities across Africa, Latin America, and Asia. In these contexts, informality, defined as unregulated commerce, adaptive reuse of space, incremental construction, and self-organized infrastructure, shapes the everyday choreography Jane Jacobs called the “sidewalk ballet.” Yet because governments rarely collect census-grade data on such activity, informality remains poorly documented and weakly understood. This thesis introduces a transferable computational framework to formalize informality by transforming street imagery into an auditable taxonomy of informal street-level elements, activities, and practices. The framework is tested in two contrasting districts, i.e. District 1 and District 5 of Ho Chi Minh City, where sidewalks are highly contested by vendors, pedestrians, and regulators. The contribution of this thesis is two-fold. First, this thesis contributes a three-stage pipeline for classifying sidewalk informality. Using Seesaw (Moll et al., 2022), a CLIP-based feedback loop retrieves and soft-labels candidate scenes. This is followed by manual verification and fine-tuning a lightweight ResNet on binary categories (e.g. stationary vs mobile vendors, etc.). Compared to the zero-shot model Qwen-VL-Max, the fine-tuned ResNet delivered more balanced performance (precision/recall: 0.62– 0.78) and better handled nuanced, context-sensitive distinctions. In contrast, Qwen-VL-Max favored recall and object salience but struggled with subtle or spatial cues like mobile vs. stationary setups. Second, this thesis also developed a taxonomy and annotated dataset of informality which was used to reveal spatial inequities in sidewalk use. By converting curbside complexity into structured, updateable categories, the framework enables planners to recognize the adaptive value of informal practices, target genuine hazards, and design interventions for more equitable urban planning.

Nuclear Microreactor-Powered Container Ships for Maritime Decarbonization

Thu, 01 May 2025 00:00:00 GMT

Nuclear Microreactor-Powered Container Ships for Maritime Decarbonization Dickerman, Matthew F. The maritime shipping industry, responsible for approximately 3% of global greenhouse gas emissions, faces growing pressure to achieve net-zero emissions by 2050 under the International Maritime Organization (IMO) framework. Alternative fuels such as liquefied natural gas, ammonia, and methanol present challenges related to energy density, infrastructure, safety, and cost. Nuclear microreactors offer high energy density, zero operational emissions, and multi-year endurance, but require coordinated regulatory development and stakeholder engagement for commercial adoption. This thesis evaluates the feasibility of integrating microreactors into container ship designs employing electric propulsion and standardized intermodal logistics. Holos-Quad microreactors are selected based on their modular architecture, transportability, and compatibility with marine operations. Detailed ship concepts are developed for Feeder, Panamax, and New-Panamax classes, accompanied by a phased fleet development strategy. Economic modeling compares the lifecycle costs of conventional and microreactor-powered ships, incorporating capital expenditures, operating costs, financing assumptions, and carbon pricing. Fleet-level analysis indicates that microreactor-powered ships can achieve comparable or improved profitability while eliminating nearly 44 million metric tons of CO2e emissions across a ten-ship fleet. Sensitivity analyses confirm the robustness of these results across a wide range of future scenarios. By integrating stakeholder analysis, technical feasibility assessments, and economic modeling, this research establishes a commercially viable framework for zero-emission nuclear-powered shipping, offering a scalable pathway toward sustainable maritime operations.

A magnetic levitation testbed for development of real-time control frameworks applied in fusion

Thu, 01 May 2025 00:00:00 GMT

A magnetic levitation testbed for development of real-time control frameworks applied in fusion Lee, Yehoon This thesis presents the development of a magnetic levitation device as a hardware-in-theloop platform to be used for research in Control and Data Acquisition frameworks applied to fusion experiments. Specifically, the testbed is aimed to demonstrate distributed, modular control using a plasma control system framework being developed at the Plasma Science and Fusion Center at MIT. This framework integrates a real-time control framework, MARTe2, and a data management framework, MDSplus, to provide platform flexibility and robust data management for rapid prototyping of control systems. Both frameworks are widely used individually in fusion experiments worldwide. The magnetic levitation setup is centered around a single electromagnet coil which levitates a permanent disk magnet from above. Implemented with the integrated MARTe2/MDSplus framework, the controller, actuator, and sensors are distributed over the network. With the magnetic levitation testbed, this thesis achieves three objectives: 1. formulation of a physicsbased model of the system, 2. development of a controller in a modular, networked framework, and 3. training and implementation of learning-based methods within the framework. First, a state-space model for single-axis magnetic levitation is formulated based on theory and refined with magnetic field measurements. A feedback controller is then developed and implemented with MATLAB Simulink. Afterwards, a vision-based observer is developed to estimate position and tilt of the levitated magnet. Pose-image datasets are auto-labeled using fiducial markers and are used to train a convolutional neural network. Finally, the trained network will be applied in system identification of the final controlled system. Through the process of system development, this thesis proposes that the integrated MARTe2/MDSplus framework is robust in performing real-time control of a networked system, and its structural modularity is advantageous for developing and testing learning-based models.

A Fast Assay of Bacteria Cell Permeability for Genetic Transformation

Thu, 01 May 2025 00:00:00 GMT

A Fast Assay of Bacteria Cell Permeability for Genetic Transformation Nieves, Charmaine Bacterial cell genetic engineering is fundamental for research aiming to learn more about bacterial species for a broad range of applications. One method of intracellular delivery of foreign DNA during the genetic engineering process is the use of electroporation to create pores along the bacterial cell membrane. Current methods for assessing pore formation do not directly measure cell permeabilization or enable same-day assessment. In this thesis, a novel fast-screening protocol combining SYTOX green, microfluidics, and fluorescence imaging is evaluated for its capability to assess multiple conditions for cell permeabilization within a single day. By imaging bulk suspensions of post-electroporated cells stained with intracellularly delivered SYTOX, multiple electroporation conditions can be rapidly screened for cell permeabilization. This fast-screening protocol utilizes standard microbiology equipment and low-cost microfluidic imaging chambers, lowering the barrier to adoption and significantly reducing experimental time compared to conventional protocols involving foreign DNA delivery. Importantly, by decoupling permeabilization assessment from foreign DNA uptake, this method isolates the effect of membrane permeabilization from confounding factors such as restriction-modification systems. As a result, it provides a more accurate qualitative and quantitative assessment of bacterial membrane disruption. This approach enables same-day evaluation of electroporation conditions regardless of bacterial growth rate, potentially accelerating the optimization process for intracellular delivery in gene editing applications.

Probabilistic Human Arm Motion Prediction via Structured Multitask Variational Gaussian Processes

Thu, 01 May 2025 00:00:00 GMT

Probabilistic Human Arm Motion Prediction via Structured Multitask Variational Gaussian Processes Chong, Jinger S. Accurate human motion prediction with uncertainty estimation is essential for safe and efficient human-robot collaboration, where robots must anticipate and react to human movements in real-time. Existing methods either rely on sophisticated techniques that demand extensive training data and sacrifice interpretability, or use simpler approaches like conventional Gaussian Processes (GPs) that fall short in performance. To address this gap, we propose a novel structured multitask variational GP framework that explicitly incorporates joint dependencies to reflect human kinematics. We further enhance this framework by integrating angular velocity constraints, which improve the physical plausibility of predictions. The addition of constraints alone yields up to a 66% reduction in mean angle error (MAE) and an 84% improvement in the likelihood of predicting ground truth (NLL), outperforming standard GP baselines across a wide range of motion types and prediction horizons. Among model variants, our structured GP with constraints offers the best tradeoff—achieving MAE within 1.1–2.6% and NLL within 0.001–0.012 of the best-performing model, while maintaining significantly lower overconfidence rates (OCR), particularly at short horizons where the independent GP model OCR reaches nearly 45%. These results underscore the importance of incorporating structure and context in human motion prediction, demonstrating that even simpler probabilistic models like GPs can achieve substantial performance gains when augmented with such information.

Permanent Magnet Synchronous Motors: Nonlinear Dynamic Modeling, Hardware Characterization, and High-Bandwidth Torque Control for Applications in Dynamic Robotics

Thu, 01 May 2025 00:00:00 GMT

Permanent Magnet Synchronous Motors: Nonlinear Dynamic Modeling, Hardware Characterization, and High-Bandwidth Torque Control for Applications in Dynamic Robotics Roy, Ronak The high-level control algorithms that are responsible for achieving dynamic locomotion in legged robots depend on accurate torque production for matching real-life performance with simulated performance. To achieve accurate torque production, actuators must run high-bandwidth, low-level torque control. Developing high performance low-level controllers requires accurate actuator models. This thesis covers the physical model of a Permanent Magnet Synchronous Motors (PMSM), a very common type of actuator in dynamic robotics. This thesis details the derivation of the PMSM linear model, how to adapt the model dependent on the physical construction of a real motor, and the implementation of FieldOriented Control (FOC) to achieve torque control. This thesis also describes a novel design of a high-precision dynamometer, which allows a motor to be coupled with an impedance and a torque sensor in order to accurately characterize the torque production characteristics of the motor. Using this dynamometer and other experimental setups, this thesis validates the model and determines parameters for multiple different actuators. Finally, this thesis proposes an augmented PMSM model that considers the nonlinear saturation behavior of the motor, validating the principle with hardware experiments, and demonstrates a nonlinear torque model and gain-scheduled current controller that improve torque tracking performance.

Development of an Apparatus and Testing Strategy for Characterizing Rolling Resistance of Omnidirectional Wheels

Thu, 01 May 2025 00:00:00 GMT

Development of an Apparatus and Testing Strategy for Characterizing Rolling Resistance of Omnidirectional Wheels Ilkbahar, Kayra B. Omnidirectional wheels (omni wheels) are a type of wheel technology similar to caster wheels but capable of simultaneous longitudinal and lateral motion, making them suitable for holonomic motion applications. In recent years, their popularity has grown substantially in areas such as educational robotics, autonomous vehicles, and industrial automation. Despite their similarity to caster wheels in both function and application, omni wheels are a much less mature technology and few agreed-upon standards exist for their design and testing. This thesis covers the design of a test procedure and its requisite test apparatus to characterize the rolling resistance of omni wheels across various test conditions, and focuses specifically on the mechanical and electrical design of an apparatus which can measure the rolling resistance coefficient of omni wheels while modulating their load weight, travel angle, and travel speed.

A Structural Approach to Measuring Time-varying Risk Aversion

Thu, 01 May 2025 00:00:00 GMT

A Structural Approach to Measuring Time-varying Risk Aversion von Turkovich, Nick Non-homothetic preferences have the potential to rationalize important asset pricing facts including time-varying risk premia and business cycle movements in asset prices (e.g., Campbell and Cochrane (1999)). This paper offers a structural approach to measuring time-varying risk aversion. Motivated by the literature on consumption commitments (e.g., Flavin and Nakagawa (2008), Chetty and Szeidl (2016), Chetty, Sandor, and Szeidl (2017)), I develop a model in which investors have nonseparable preferences over housing and nonhousing consumption, and investors must consume a minimum amount of housing each period. Non-housing consumption is assumed to be flexibly chosen. The key insight is that the intratemporal optimality condition between the two goods reveals information about the surplus consumption ratio, a key variable driving risk aversion. A cointegrating relationship between relative quantities and prices allow us to identify the elasticity of intratemporal substitution and measure surplus housing consumption. Using aggregate U.S. consumption data from 1959 to the present, the measured surplus consumption ratio demonstrates clear business cycle fluctuations, rising during expansions and falling during recessions. Consistent with the theory, this measure also predicts future excess returns.

Decarbonization of Gas Heating in Massachusetts: An Evaluation of Current Trends and Opportunities

Thu, 01 May 2025 00:00:00 GMT

Decarbonization of Gas Heating in Massachusetts: An Evaluation of Current Trends and Opportunities Epstein, Andrew The commonwealth of Massachusetts has ambitious decarbonization goals enshrined in law and has been establishing the regulations to achieve them. Through its Department of Public Utilities regulatory rulings, the state has required local gas and electric utilities to pursue decarbonization not only by reducing the emissions of their electric supply but also by actively supporting gas load reduction. The residential heating sector dominates this effort, with programs like MassSave incentivizing customer adoption and now MA DPU 20-80-B requiring gas utilities to demonstrate that they have sufficiently evaluated the possibility of non-pipeline alternatives, including but not limited to electrifying customers instead of reinvesting in the gas system for all future gas investments. This paper looks at a single Massachusetts utility, National Grid, and evaluates where its customers are switching to electric heat and which mechanisms are driving current adoption. It further evaluates where geographically National Grid could invest in electrification instead of replacing gas investments under the new 20-80-B order. In doing so it establishes a model for cost benefit calculations related to prospective NPA projects. This paper then examines the degree to which ongoing electrification efforts are aligned with one another. Finally, this paper explores concerns that the process of electrification might be regressive, leaving behind those who cannot afford to electrify their systems and leaving them to pay ever-increasing prices as the full gas system is paid for through rates from a decreasing population of consumers. In evaluation of such concerns, it determines the geographic correlation between ongoing decarbonization efforts and communities already facing housing burden.

Streamlining Diagnostics of Electrical-Connection-Related Errors in General Assembly Using Augmented Reality Wearables

Thu, 01 May 2025 00:00:00 GMT

Streamlining Diagnostics of Electrical-Connection-Related Errors in General Assembly Using Augmented Reality Wearables Salata, Elizabeth Electrical connection errors arise frequently during manufacturing. It is optimal to repair these errors during General Assembly Trim Line stations when the wiring harnesses are still exposed and easily accessible. However, the time required to locate the cause of the errors often exceeds Trim station cycle times, so most repairs are delayed until after General Assembly. Due to the implications of shutting down the line, this results in significantly higher repair times, scrap costs, and resources. To overcome these challenges, there is clear evidence supporting the use of Augmented Reality (AR) tools to innovate and streamline manufacturing processes. This master's thesis identified deficiencies in the current standard operating procedure for addressing errors and used a human-centered design approach to develop a novel error diagnostic process using an AR overlay technique to pin point on the vehicle where the problem lies. This thesis also conducted an experiment to assess the performance, success rate, and perceived cognitive load of the two processes. The data collected from the experiment provided sufficient evidence that the diagnostic process developed for this thesis reduces the elapsed time to locate the connection error by 75% with a statistically significant reduction in overall perceived cognitive load. The likelihood of widespread adoption of the AR overlay process was assessed from an estimate of further AR hardware development, safety considerations in automotive manufacturing environments, and the level of enthusiasm of all stakeholders who were consulted for this research project.

A Techno-Economic Assessment of Hybrid Renewable Energy and Battery Storage Systems for Data Centers

Thu, 01 May 2025 00:00:00 GMT

A Techno-Economic Assessment of Hybrid Renewable Energy and Battery Storage Systems for Data Centers Sirgo, Alex As the demand for data centers continues to grow, so does their energy consumption, making it increasingly important to develop sustainable and cost-effective strategies for powering them with carbon-free electricity. This thesis explores a techno-economic modeling framework that evaluates combinations of solar, wind, and battery energy storage systems to assess their ability to meet a data center’s electricity demand with on-site renewable generation. The model fills a gap in current literature by focusing on real-time energy matching using co-located infrastructure, rather than traditional off-site procurement methods like power purchase agreements and renewable energy credits. Using real-world weather and price data, the simulation calculates hourly generation, storage behavior, and grid interactions across a 20-year period. A financial model then calculates the levelized cost of energy (LCOE) for each system configuration. Results show that wind energy generally provides the lowest-cost renewable supply option, while hybrid solar and wind configurations improve renewable penetration. Battery storage plays a key role in shifting excess generation to periods of undersupply, but its economic viability depends on system sizing. Across different system configurations, renewable penetration ranged from 31.3% to 97.8%, while LCOE varied from $27.5/MWh to over $100/MWh, illustrating the trade-offs between cost and grid independence. By providing a structured analysis of the trade-offs between renewable penetration and cost, this research offers insight into how data centers and other energy-intensive facilities can design dedicated carbon-free energy systems. The findings underscore the importance of balancing resource diversity and storage investment to achieve decarbonization goals while maintaining economic viability.

Diagnostics in Additive Manufacturing Using Image-Based Machine Learning

Thu, 01 May 2025 00:00:00 GMT

Diagnostics in Additive Manufacturing Using Image-Based Machine Learning Varma, Arun Alejandro Additive Manufacturing (AM) is a vital capability in the aerospace industry. Blue Origin manufactures a substantial share of engine parts via metal AM. To meet growing customer demand, the company must dramatically increase engine throughput and, thus, 3D prints. Blue Origin has identified non-destructive testing (NDT) – particularly, Computed Tomography (CT) scanning – as an unsustainable bottleneck to expanding AM capacity. Not only is this process expensive, but, critically, there are not enough aerospace-grade CT machines in the world to support projected throughput. Without process change, meeting customer demand will soon become impossible. Yet, these scans provide important quality control, and any reduction in NDT must be accompanied by assurances of engine part integrity. This thesis introduces a diagnostic system that safely alleviates the bottleneck, and further yields insights that end-stage NDT alone cannot provide. The proposal is a machine learning system that evaluates the manufacturing process itself, examining layer-by-layer photographs captured during printing. It is predicated on two hypotheses: (1) These images, considered together, provide a synthetic 3D illustration of the build process; and (2) Machines can be taught to assess these process signatures dependably. The resulting system provides rich diagnostics. It achieves near-perfect anomaly recognition – 100% when using conservative defect thresholds. Operationally, the system can (at minimum) safely enable a 37-54% reduction in NDT, translating to millions of dollars in annual cost savings. In practice, this reduction will likely be higher. The system further enables early process intervention and a more data-driven approach to manufacturing intelligence. This work turns what began as an unsustainable bottleneck into an opportunity for enhanced quality control, process intelligence, and long-term manufacturing resilience.

Substitution among Social Media Platforms: Evidence from App Tracking Panel Data

Thu, 01 May 2025 00:00:00 GMT

Substitution among Social Media Platforms: Evidence from App Tracking Panel Data Lagutina, Rina This thesis explores a novel approach to competitive intelligence in the social media ecosystem by leveraging external mobile panel data to study substitution dynamics. It focuses on contextspecific behavioral patterns to identify which platforms compete for user attention in given situations. Using mobile app session data from April 2023 for approximately 5,000 users, the analysis segments usage into three behavioral contexts – morning, evening, and at-home sessions – and characterizes user-app interactions through descriptive statistics. K-means clustering is applied to identify archetypes of usage behavior across these contexts, revealing distinct patterns such as quick-check habits, deep content consumption, and intensive texting. By comparing app usage profiles across contexts, the study uncovers shifts in how and when platforms are used, highlighting subtle substitution dynamics. To validate the findings, the study analyzes app usage during service outages, testing if potential substitutes see increased engagement when a competing platform is unavailable. These insights offer a richer, contextaware framework for product managers to uncover indirect competition and tailor platform strategies to specific user behaviors. Limitations include reliance on behavioral data without content-level detail, mobile-only focus, and demographic skew in the panel.

Harnessing Generative AI in Developing Economies: A Systems Framework for Policy Design in Bangladesh

Thu, 01 May 2025 00:00:00 GMT

Harnessing Generative AI in Developing Economies: A Systems Framework for Policy Design in Bangladesh Bari, Md Mustabeen Ul This thesis develops a systems-based policy framework for Generative Artificial Intelligence (GenAI) implementation in developing economies, with specific application to Bangladesh. While GenAI's potential productivity and labor market impacts are well-studied in developed economies, limited research addresses the challenges faced by developing countries positioned primarily as technology consumers rather than producers. The research employs causal loop diagramming to map interactions between five critical policy domains: human capital development, digital infrastructure, data sovereignty, sectoral stimulus, and governance. The resulting framework identifies four primary reinforcing mechanisms that can accelerate adoption and three balancing mechanisms related to labor displacement. To validate the framework, the research analyzes contrasting implementation approaches from India and Egypt, demonstrating the importance of cross-domain synergies in effective policy design. Applied to Bangladesh, the framework yields a dual-entry strategy focusing on healthcare and education sectors as initial implementation domains, leveraging the country's strategic advantages while addressing resource constraints through a consortia-based implementation model that creates institutional resilience. The thesis contributes both a reusable conceptual toolkit for analyzing GenAI policy in resource-constrained settings and an initial context-anchored roadmap for Bangladesh. Future research should refine the framework through longitudinal case studies while developing more detailed, stakeholder-engaged implementation plans for Bangladesh that include concrete budget allocations, institutional responsibilities, and measurable outcomes.

The Value of Digitizing Manufacturing Environments

Thu, 01 May 2025 00:00:00 GMT

The Value of Digitizing Manufacturing Environments Briggi, Conor S. There is significant variability and dispute around the value of digitally transformed manufacturing environments and no single methodology is broadly accepted. The variability stems from time-dependencies, implementation effectiveness, and the dynamic environments digital solutions are deployed in. However, an accurate accounting of this value is essential to company strategic planning. The research outlines how to approach this variability, cost parameters to consider, primary sources of value generation, and best practices for implementing Smart Factories. A tool that addresses these issues was successfully developed and deployed at Stanley Black & Decker, helping the company to assess performance of the digitization efforts and tailor the delivered solution to optimize manufacturing performance. Results from this tool showed a positive expected return on investment and are provided to contextualize efforts in similar areas.

Multimodal Generative AI Chatbot for Root Cause Diagnosis in Predictive Maintenance

Thu, 01 May 2025 00:00:00 GMT

Multimodal Generative AI Chatbot for Root Cause Diagnosis in Predictive Maintenance Lorente Anon, Carla Predictive maintenance plays a critical role in industrial operations by enabling organizations to detect potential equipment failures before they occur. However, while sensor data can identify anomalies such as excessive vibration or temperature fluctuations, technicians often struggle to efficiently diagnose and resolve the root causes of these alarms. This research presents a generative AI-powered chatbot designed to enhance the root cause diagnosis process in predictive maintenance by leveraging multimodal retrieval-augmented generation (RAG) and advanced AI-driven troubleshooting capabilities. The chatbot integrates multiple functionalities to support maintenance teams in resolving alarms quickly and accurately. Its time series analysis module processes real-time sensor data, identifying abnormal patterns and guiding users through a structured troubleshooting workflow. The retrieval-augmented generation (RAG) engine allows the chatbot to retrieve and synthesize relevant troubleshooting information from technical manuals, historical maintenance records, and structured knowledge bases, ensuring that technicians receive precise, grounded outputs. Additionally, the chatbot supports multimodal interactions, enabling users to upload images, audio, and video for more comprehensive diagnostics. By analyzing uploaded images of damaged components, transcribing spoken maintenance reports, and processing video footage of equipment malfunctions, the chatbot enhances problem identification and resolution. Another key feature of the chatbot is its interactive guided conversation system, which enables multi-turn dialogues that refine diagnostics dynamically based on technician input. Instead of providing static troubleshooting steps, the chatbot continuously adapts its responses to ensure that users receive the most relevant recommendations as the diagnostic process unfolds. To maintain safety and reliability, the system incorporates AI guardrails, filtering inappropriate or irrelevant inputs while ensuring that generated responses align with best practices for industrial maintenance. An evaluation framework is proposed to assess the chatbot’s effectiveness, focusing on retrieval accuracy, response relevance, and diagnostic efficiency. Initial results demonstrate approximately 30% reduction in diagnostic time, highlighting the chatbot’s potential to improve maintenance workflows, reduce downtime, and enhance technician productivity. This research underscores the transformative role of multimodal generative AI in predictive maintenance and lays the foundation for broader industrial applications. As a result of this work, a patent has been filed to protect the novel architecture and methods developed. Future work could focus on expanding retrieval capabilities to include video, integrating intelligent task automation for dynamic work order generation, and refining alarm prioritization using adaptive risk-based assessments.

Quantifying over Individual Concepts

Fri, 01 Sep 2023 00:00:00 GMT

Quantifying over Individual Concepts Kobayashi, Filipe Hisao Since Montague (1973), it has been assumed that quantificational DPs must, at least sometimes, be analyzed as quantifiers over individual concepts (i.e., functions from indices of evaluation to individuals). Because the domain of individual concepts is significantly greater than that of individuals, the challenge has always been how to properly constrain quantification over these objects. This dissertation proposes a solution to this problem by developing a novel theory as to how NPs are shifted from predicates of individual into predicates of individual concepts. The idea is that, since NPs are interpreted as restrictors, the nature of this shifting mechanism will constrain quantification. The proposal bears a strong resemblance to the analysis of interrogative clauses of Karttunen (1977): suitable predicates of individual concepts are built from the interaction of a type-shifting operation and existential quantifiers. In three cases studies, I show how this theory can solve old and new puzzles: (i) the different interpretations of sentences of the form ‘[Det NP] changed’ (Nathan 2006); (ii) two ambiguities in the interpretation of concealed questions (Heim 1979); and (iii) question intruders, a novel puzzle concerning the interpretation of both embedded interrogative clauses and concealed questions.

Within ‘Reason’: A Study of Normative Language

Fri, 01 Sep 2023 00:00:00 GMT

Within ‘Reason’: A Study of Normative Language Watkins, Eliot What do we mean when we say that someone ought to do something? What do we mean when we say that someone has a reason to do something? What do we mean when we say that someone has more reason to do one thing rather than another? The primary goal of this project is to shed light on these semantic questions. The picture of normative talk that I develop across this thesis has a distinctive feature: the notion of a reason (roughly, a fact that counts in favour of something) isn’t given any fundamental role to play. Instead, the meanings of ‘ought’, ‘must’ and ‘is a reason for…’ are all understood in terms of something gradable – they’re understood in terms of facts about how much reason there is for something to be done. Chapter One focuses on deontic modals like ‘ought’ and ‘must’. I argue that the standard semantics for these expressions is incompatible with the idea that facts about what you ought to do are connected with facts about what you have reason to do. I develop a new semantics for deontic modals which builds-in the connections between ought and reasons from the ground up. Chapter Two centres on ‘reason’. We use ‘reason’ as both a count noun (as in “there is a reason for you to read my dissertation”) and a mass noun (as in “there is some reason for you to read my dissertation”). I argue that the best semantics for ‘reason’ will treat the mass form as fundamental. ‘Reason’ is a predicate of a particular kind of state – the state someone is in when they have reason to do something. I turn this result into an argument against the enduringly popular idea that count noun reasons are normatively fundamental. Chapter Three stays with reasons. According to a standard picture, normative reasons do not extend beyond the boundaries of agency. If something isn’t an agent – if it can’t do rudimentary practical reasoning – then there can’t be normative reasons for it to do one thing rather than another. I argue that this standard picture gets things totally wrong: there are reasons for non-agents to be certain ways and do certain things. We must not analyse what it is to be a reason by appealing to distinctively agential capacities.

Lessons from CP in Passamaquoddy and beyond

Fri, 01 Sep 2023 00:00:00 GMT

Lessons from CP in Passamaquoddy and beyond Grishin, Peter Nicholas This thesis explores various aspects of CP morphosyntax in Passamaquoddy-Wolastoqey and other Algonquian languages and their consequences for broader generative syntactic theory. It consists of two parts: one investigates clause typing and clause size in Passamaquoddy, and the other investigates the properties of a CP-layer agreement marker, the peripheral suffix, across Algonquian. In addition, a lengthy background chapter offers new data and insight on the correct analysis of the inverse and obviation in Passamaquoddy and across Algonquian. Part I studies the distribution of the three morphologically-distinguished non-imperative clause types in Passamaquoddy: the independent, the conjunct, and the subordinative. I argue that their distribution in complementation and coordination structures falls out naturally from their structural size, following the work of Wurmbrand and Lohninger (2023) and Bjorkman (2012, 2013). I support this conclusion by carefully investigating how each clause type interacts with Ā phenomena like wh movement and long distance agreement, showing that various complex interactions between these syntactic processes are derivative of clause size: independent clauses and conjunct clauses under epistemic attitudes are large, phasal CPs, conjunct clauses under direct perception predicates are smaller, non-phasal CPs, and subordinative clauses are bare TPs. Part II studies two unexpected properties of peripheral agreement across Algonquian: (i) its preference for agreeing with third persons, no matter their syntactic role (found in all Algonquian languages); and (ii) its preference for agreeing with the least local goal (found in languages like Passamaquoddy, Ojibwe, and Wampanoag). I explore the consequences of these typologically unusual properties for the theory of φ agreement and provide an analysis of the cross-Algonquian variation we find in peripheral agreement (building on Xu 2021, 2022). I argue that Algonquian third person preference forces us to accept Nevins (2007) and Trommer’s (2008) conclusion that third person cannot be underspecified relative to first and second person, even in the syntax (contra Preminger 2019a and van Alem 2023). Additionally, I show that Algonquian lowest preference doesn’t force us to give up on standard locality properties of Agree, and argue for an analysis under which C agrees with all matching accessible goals, but only spells out the last Agree relation—Expone Outermost—building a parallel with similar ideas in the domain of multiple case assignment. Finally, I capture cross-Algonquian variation in peripheral agreement by varying the specification of the peripheral agreement probe and varying which arguments are able to shift out of the VP phase.

*ABA in Multidimensional Paradigms: A MAX/DEP-based account

Fri, 01 Sep 2023 00:00:00 GMT

*ABA in Multidimensional Paradigms: A MAX/DEP-based account Zompì, Stanislao The last decade and a half has witnessed intensive research into *ABA universals—generalizations such as “If a nominative and the corresponding dative have the same exponent, then the corresponding accusative has that exponent, too” (Caha 2009; Smith et al. 2019). Most existing work on these universals has only focused on one ‘paradigm column’ at a time, by checking a given paradigm’s nominative singular, accusative singular, and dative singular, for example, with no heed to whether any of the relevant exponents would also show up in that paradigm’s nominative plural, accusative plural, or dative plural. However, some recent literature has pointed out that inspecting full paradigms is crucial to our understanding of *ABA, because some classic accounts that derive *ABA column-internally turn out to also make predictions as to what may or may not happen across columns, and those predictions are often incorrect (cf., among others, Christopoulos & Zompì 2022). In this dissertation, I review those incorrect predictions and replace them with a novel generalization specifically concerning *ABA-like effects in multidimensional paradigms. I then set out to derive this generalization by setting up an exponent-selection system wherein exponents may both be underspecified and be overspecified with respect to their exponenda, with each of these departures from a perfect match being penalized but not necessarily fatal. In particular, I explicitly implement this intuition in optimality-theoretic terms, via a strict-domination ranking of violable Max and Dep constraints (cf. in particular Ackema & Neeleman 2005; Wolf 2008; Müller 2020), and I show that the resulting system, while restrictive enough to derive the desired generalization, is also powerful enough to afford a natural account of some notoriously unnatural (‘morphomic’) exponent distributions in the inflection of Germanic pronouns and Romance verbs.

Intangible Investments and the Accrual-Cash Flow Relationship

Thu, 01 May 2025 00:00:00 GMT

Intangible Investments and the Accrual-Cash Flow Relationship Soares, Fabio This paper investigates whether the weakening negative relationship between accruals and operating cash flows can be attributed to the immediate expensing of intangible investments under current accounting standards. Building on the framework proposed by Green et al. (2022), I examine how the mechanical capitalization of intangible investments affects the accrual-cash flow relationship across firms with varying R&D intensities. I show that the capitalization impacts the relationship in unexpected ways, indicating that the proposed rationale cannot fully explain the observed trend. I further exploit differences in accounting treatments under IFRS and US GAAP to test whether increased capitalization of intangible investments through development costs strengthens the relationship. I find that the relationship is significantly more negative under IFRS than US GAAP, independently of R&D expenditure, suggesting that increased capitalization alone does not explain the differences. Additionally, the positive trend observed for high R&D firms in both standards highlights that increased capitalization is insufficient to reverse the weakening trend. These results challenge the view that current accounting practices are the primary cause of the weakening accrual-cash flow relationship and underscore the need for further exploration of alternative explanations.

Winning Over Gen Z: The Evolving Strategies of Sports Leagues and Media in Response to Changing Youth Habits

Thu, 01 May 2025 00:00:00 GMT

Winning Over Gen Z: The Evolving Strategies of Sports Leagues and Media in Response to Changing Youth Habits Zeng, Arnaud This thesis examines how sports leagues and media companies are evolving to better connect with Generation Z, a generation whose changing expectations and habits – on-demand and socially driven – are reshaping the landscape of sports consumption. With fewer Gen Z fans watching full games on traditional mediums, the industry is being pushed to rethink its approach, adapting not just how content is delivered, but also what kind of content is created. Through a combination of expert interviews and industry data, this paper looks at the rise of short-form content, the importance of digital-first platforms, and the growing influence of storytelling through influencers or behind the scenes. It also explores how new competition formats are exploiting what it now means to be a fan. The goal is to understand how the sports ecosystem is adjusting to remain relevant to its youngest audience.

Metal Additive Manufacturing Capabilities for Footwear Prototyping and Product Creation

Thu, 01 May 2025 00:00:00 GMT

Metal Additive Manufacturing Capabilities for Footwear Prototyping and Product Creation Xi, Tiffany In the footwear industry, the speed in which footwear designs reach the market impacts the ability for a company to accurately meet the demands of its customers as the probability of consumer preferences changing increases with time. This research investigates the impact of incorporating metal additive manufacturing capabilities into the product creation process of a major athletic footwear company. The study aims to determine whether and under which applications metal additive manufacturing can increase the speed at which footwear designs reach the market, while maintaining or improving the desired product quality. A case study approach was employed, focusing on the development of rubber outsole molds using metal additive manufacturing technology. The study compared two process flows that excluded and included metal additive manufacturing. The case study evaluated these processes based on the speed of the development process and the quality of the produced footwear samples. The footwear sample quality was measured against production-equivalent samples obtained from the company’s manufacturing partner. The results demonstrated that incorporating metal additive manufacturing capabilities led to a reduction in the time required for mold design and fabrication. This speed advantage was primarily attributed to the ability to directly fabricate detailed textures into the mold, eliminating the need for outsourced etching processes. The visual quality of samples produced did not fully match those created by the company's manufacturing partners but were sufficient for initial sample development. Importantly, the traction properties were comparable to those of the manufacturing partner's samples, indicating that the functional quality of the samples is adequate for product development purposes. This research provides valuable insights into the potential of metal additive manufacturing in accelerating footwear product development. Future work recommendations include exploring advanced modeling and design software and examining the impact of machine parameters on build quality. The findings of this study have implications for both the footwear industry and other sectors considering the integration of metal additive manufacturing technologies into their product development processes.

Evaluating Impact Investing through a Systems Thinking Lens: Hallmarks of a Transformational Approach

Thu, 01 May 2025 00:00:00 GMT

Evaluating Impact Investing through a Systems Thinking Lens: Hallmarks of a Transformational Approach Zhang, Yu (Sherry) As impact investing increasingly aspires to drive systemic change, the question of how to evaluate such efforts remains underexplored. Traditional evaluation approaches often grounded in linear causality and program-level outputs, and struggled to capture the complexity, interdependence, and emergent nature of systemic transformation. This thesis investigates how systemic investing can be evaluated by integrating systems thinking, evaluation theory, and investing practice. It develops a conceptual framework of thirteen hallmarks that characterize systemic investing evaluation across dimensions such as time horizons, stakeholder engagement, cross-sector collaboration, and capital dynamics. Drawing on 46 real-world cases, the research identifies 112 indicators to make these hallmarks observable and assessable in practice. To support practical application, the thesis also introduces an AI-assisted scoring tool that automates the evaluation of narrative content using the framework. Together, these contributions aim to support more reflective, adaptive, and system-aware evaluation practices in the emerging field of systemic investing.

Multi-Objective Optimization of Container Load Plans for Modulating Inventory Flow

Thu, 01 May 2025 00:00:00 GMT

Multi-Objective Optimization of Container Load Plans for Modulating Inventory Flow Sen, Shweta Conventional strategies for container load planning (CLP) predominantly emphasize maximizing container utilization, which can result in suboptimally-timed inventory arrival, increased inventory holding costs, and downstream operational inefficiencies. Using a real-world case study from a global footwear and apparel retailer, this research formulates a novel multi-objective mixed-integer linear programming (MOMILP) model that jointly considers container utilization, transportation and storage costs, and timing accuracy of inventory delivery. The proposed model utilizes a branch-and-bound algorithm to evaluate numerous load configurations, assessing the impact of different load rules and weighting parameters on transportation performance metrics and inventory flow. Results highlight the cruciality of prioritizing delivery precision in transportation management decisions, demonstrating that solely maximizing volume utilization can adversely affect overall cost efficiency when downstream inventory storage and operational requirements are considered. This work also provides a process map of load planning activities and identifies targeted operational improvements, such as consolidation bypass and purchase order (PO) partitioning, that can enhance inventory flow smoothness, reduce transportation costs, and support more responsive logistics networks. Collectively, this work extends existing CLP methodologies by incorporating delivery timing and inventory storage considerations into load planning decisions, offering practical enhancements for logistics optimization.

Principles and Practices of Gap-Closing Investing

Thu, 01 May 2025 00:00:00 GMT

Principles and Practices of Gap-Closing Investing Kapor, Mitchell This thesis examines the principles and practices of gap-closing investing, a distinctive model of early-stage venture capital investing that seeks to close gaps in access, opportunity, and outcomes for low-income communities and communities of color. Developed by Dr. Freada Kapor Klein and Mitchell Kapor through Kapor Capital, gap-closing investing integrates social impact objectives with a performance-driven investment strategy. The thesis combines historical analysis of socially responsible investing and impact investing with case studies of venturebacked startups to situate gap-closing investing within a broader tradition of values-based finance. It traces the ethical roots of impact investing to religious traditions, the emergence of socially responsible investing funds in the 1970s, and the formalization of impact investing terminology in the late 2000s. Gap-closing investing is distinguished by a developmental approach to startup growth, a redefinition of founder selection criteria emphasizing “distance traveled” over pedigree, and a focus on mitigating structural barriers through capital allocation. The thesis critically compares gap-closing investing to Corporate Social Responsibility (CSR) and Environmental, Social, and Governance (ESG) frameworks, arguing that gap-closing uniquely centers systemic impact as a core investment goal rather than a secondary consideration. The findings challenge the perception that impact investing is inherently concessionary, using performance data from Kapor Capital’s portfolio to demonstrate that intentional, equity-focused investing can produce both superior financial returns and measurable social outcomes. Gap-closing investing is presented as both a pragmatic investment strategy and a model for using venture capital to drive systemic change toward a more inclusive economy.

Predictive Model for Battery State of Health

Thu, 01 May 2025 00:00:00 GMT

Predictive Model for Battery State of Health Garza Lozano, Catalina As battery energy storage systems (BESS) become critical components of grid infrastructure, accurately assessing their State of Health (SoH) is essential for optimizing performance, reducing costs, and ensuring contractual compliance. This thesis investigates the development of accurate, real-time SoH estimation models for utility-scale battery storage sites operated by NextEra Energy. Current SoH measurements—derived from annual capacity tests and Battery Management System (BMS) data—are often inaccurate or infrequent, leading to either over- or under-augmentation and resulting in financial inefficiencies. To address this gap, four state estimation models were developed and evaluated: an Unscented Kalman Filter (UKF), a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN), a multitask RNN, and a Delayed Reinforcement Learning (DRL) model. Each model uses operational data—such as voltage, current, temperature, and State of Charge (SoC)—to estimate degradation patterns and predict SoH at the rack, lineup, and site levels. Their outputs were compared against ground-truth capacity test results from a large-scale battery storage site. The DRL model demonstrated the highest accuracy, achieving a deviation of only 1.6 months compared to capacity test data, significantly outperforming existing BMS readings and the other three models. These findings underscore the value of advanced machine learning techniques in enabling proactive maintenance, optimized augmentation scheduling, and cost-efficient storage site management. This research offers a scalable framework for real-time SoH estimation across large fleets of battery storage assets and contributes to the broader goal of improving grid reliability through smarter energy storage management.

Data-Driven Key Performance Indicator Modeling for Robotic Mobile Fulfillment Systems

Thu, 01 May 2025 00:00:00 GMT

Data-Driven Key Performance Indicator Modeling for Robotic Mobile Fulfillment Systems Sowards, Steffan This work presents a study on the development and application of data-driven operational efficiency and throughput Key Performance Indicator (KPI) modeling for Robotic Mobile Fulfillment Systems (RMFS). Through rigorous analysis of extensive operational data from an operating RMFS, we demonstrate the efficacy of machine learning approaches in predicting and optimizing the performance of complex warehouse automation systems. The research employs advanced techniques, including gradient boosted bagged tree ensembles and AutoML, to capture complex input interactions and provide parallel predictions across multiple KPIs. Our models achieve a mean R² value of 0.7838 across all templates and KPIs, with particularly strong performance in our top performing metric across templates (mean R² of 0.9660). The study introduces a novel framework for feature engineering and selection, emphasizing actionable inputs while excluding intermediate variables to enhance model interpretability and practical utility. We validate our approach against novel operating conditions, demonstrating the models’ ability to generalize to unseen scenarios. Interpretability techniques, including SHAP analysis and permutation feature importance, provide valuable insights into system behavior and key performance drivers. This research establishes a generalizable framework for leveraging data-driven modeling in predicting and optimizing brownfield warehouse automation system behavior. The developed approach offers significant potential for enhancing operational decision-making, system design, and strategic planning in the rapidly evolving field of e-commerce fulfillment.

Essays on Bayesian Entrepreneurship: Evaluating and Commercializing Unconventional Ideas

Thu, 01 May 2025 00:00:00 GMT

Essays on Bayesian Entrepreneurship: Evaluating and Commercializing Unconventional Ideas Gius, Luca This dissertation investigates a fundamental challenge complicating the evaluation and commercialization of entrepreneurial opportunities: some ideas are valuable precisely because not everyone recognizes their worth. The first essay analyzes barriers against the commercialization of contrarian ideas. Researchers working with unpopular AI algorithms tend to commercialize their work only after a successful public evaluation. Those who clear this hurdle subsequently achieve better entrepreneurial outcomes. A regression-discontinuity analysis shows that this partly reflects status quo bias: for unpopular methods only, winning a contest serves as a certification, channeling disproportionate resources to the winner while equally strong near-misses remain sidelined. The second essay finds that greater judge disagreement in venture competitions predicts higher future success, especially for more distinctive startups. The third essay shows that skewness in idea value exacerbates asymmetric information in markets for ideas. Using data from auctions for digital businesses, I illustrate how this can explain why online marketplaces for ideas have struggled to emerge despite lowering transaction costs: informational frictions severely depress bids and prevent high-value digital startups from trading. The final essay, coauthored with Alfonso Gambardella and Scott Stern, introduces the archetype of Homo Entrepreneuricus: an entrepreneur who deliberately tests subjective beliefs through structured experimentation to navigate uncertainty.

Exploiting Additive Structure in Algorithm Design and Fine-Grained Complexity

Thu, 01 May 2025 00:00:00 GMT

Exploiting Additive Structure in Algorithm Design and Fine-Grained Complexity Jin, Ce In this thesis, we investigate the fine-grained complexity of various algorithmic problems with an additive flavor, including 3SUM, Subset Sum, and their close relatives. We explore their connections to various areas, such as graph algorithms, discrete optimization, combinatorial pattern matching, and computational geometry. Our new results include improved algorithms and conditional lower bounds for a wide range of problems, answering multiple open questions from the literature: • Conditional lower bounds for graph problems: We prove new lower bounds for 4-Cycle Listing and Approximate Distance Oracles conditioned on the 3SUM Hypothesis. As a key intermediate step, we show a fine-grained reduction from 3SUM to the special case of 3SUM where all pairwise sums of input numbers are distinct. • Combinatorial pattern matching: We design improved algorithms for Text-to-Pattern Hamming Distances, Pattern Matching with Wildcards, and Geometric Pattern Matching, by drawing connections from 3SUM and sparse convolution. • Knapsack-type problems: We obtain a pseudo-polynomial time algorithm for 0-1 Knapsack with (conditionally) near-optimal dependence on the maximum item weight, an improved approximation scheme for the counting problem #Knapsack, and improved exponential time algorithms for the total search problem Pigeonhole Equal Subset Sum. In order to obtain these results, we employ and develop techniques based on convolution algorithms and their extensions, as well as classic tools from additive combinatorics.

Systems Approach to Component Code Optimization for Wound Closure Portfolio

Thu, 01 May 2025 00:00:00 GMT

Systems Approach to Component Code Optimization for Wound Closure Portfolio Dubelier, Madeline Product portfolio management involves strategically analyzing, optimizing, and expanding a company’s offerings to maximize value and align with business goals. While companies often focus on portfolio expansion to meet evolving customer needs and gain market share, product deletion is frequently overlooked, leading to code proliferation and undermining operational efficiency. Effective variety management often requires input from stakeholders across the supply chain, yet few published methods take this approach. This work presents a systematic supply chain management approach to portfolio optimization using a case study from Johnson & Johnson MedTech. The case study is on pledgets, key components in non-absorbable suture systems. Recent pledget product quality issues exposed the need for a systematic approach to reducing component variety and operational efficiency. A current-state analysis addressed multiple dimensions of complexity. The evaluation combined qualitative and quantitative data and led to a five-stage optimization strategy. The proposed future state portfolio reduces component variety by 60%, guided by three constraints: continue to meet customer needs, protect competitiveness, and reduce manufacturing complexity. This method provides a replicable model for rationalizing legacy portfolios in the medical device industry.

Content Creator Conduct

Thu, 01 May 2025 00:00:00 GMT

Content Creator Conduct Du, Jason This thesis investigates the behaviors of content creators. The first study examines whether musicians learn from the success of earlier songs when they create new ones, finding that tracks on a musician’s next album tend to be more similar to the songs that performed better on their current album. The second study explores the cultural, social, and psychological aspects of content creation by tracing first-person singular pronoun usage in contemporary music, revealing geographic, temporal, and genre-based patterns. The third study analyzes the association between content creators' learning tendencies and the explainability of previous outcomes, showing that news editors are more likely to resemble previous popular headlines when those outcomes are more explainable. Collectively, these studies facilitate understanding of the factors that underlie content creation.

An Optimization-Based Approach to Efficient Clearance Inventory Allocation

Thu, 01 May 2025 00:00:00 GMT

An Optimization-Based Approach to Efficient Clearance Inventory Allocation Perez Munoz, Karla Mayra Allocating clearance inventory effectively remains a critical challenge in retail environments characterized by short decision cycles, fluctuating demand, and operational constraints. Decisions made during the clearance period are particularly impactful, as they determine the final opportunity to recover value from unsold products before they lose relevance or perishability. This thesis presents a mathematical optimization model designed to support the redistribution of discounted articles across a network of stores, with the objective of maximizing revenue while satisfying constraints related to stock availability, store capacity, and observed demand at the article-size level.Developed in collaboration with a leading global fashion retail company, the model was built to align with existing business processes and balances analytical rigor with simplicity in implementation. The model incorporates business-defined parameters and is tested using real operational data from selected distribution centers. It demonstrates significant improvements over the current practice of single-item allocation and addresses the computational challenges posed by the high dimensionality of real-world retail problems. By implementing efficient iterative procedures and demand-scaling mechanisms, the model ensures tractability while capturing the complexity of the business environment.

Gas Network Preparations for Networked Geothermal

Thu, 01 May 2025 00:00:00 GMT

Gas Network Preparations for Networked Geothermal Serbent, M. Patrick As Massachusetts pursues its goal of achieving net-zero carbon emissions by 2050, the transition from natural gas to sustainable thermal energy solutions presents both opportunities and challenges for its 1.6 million natural gas customers. This thesis investigates the potential of networked geothermal systems as a viable alternative to traditional natural gas infrastructure, with a focus on leveraging existing gas network replacement programs, such as the Gas System Enhancement Plan (GSEP), to facilitate this shift. A four-phase methodology —encompassing site selection, model development, cost analysis, and business case formulation— evaluates the feasibility of integrating high-density polyethylene (HDPE) piping into leak-prone pipe replacement efforts as a preparatory step for future geothermal or hydrogen applications. Findings suggest that HDPE offers potential material and inventory cost advantages over medium-density polyethylene (MDPE), with added flexibility for low-carbon conversions, though significant upfront costs and regulatory uncertainties remain barriers. An example site already scheduled for main replacement work showed a 6% total increase in cost for the project based on the change in pipe from MDPE to HDPE. This work underscores the potential of aligning infrastructure modernization with climate goals, offering a framework for utilities like National Grid to navigate the energy transition in cold, densely populated regions.

Machine Learning and Biosecurity in the Age of Pandemics: Advancing Biological Research and Safeguarding Synthetic Biology

Thu, 01 May 2025 00:00:00 GMT

Machine Learning and Biosecurity in the Age of Pandemics: Advancing Biological Research and Safeguarding Synthetic Biology Siddiqui, Sameed Muneeb This thesis explores the dual imperatives of enhancing biosecurity and accelerating outbreak response. The research addresses two key areas. First, the thesis analyzes the implications of a national nucleic acid synthesis screening framework on outbreak response agility. A first-hand perspective is provided, identifying potential bottlenecks stemming from lagging customer verification and sequence screening approaches. Concrete solutions, such as pre-verification of first responders, priority processing channels, pre-approval of standard countermeasure sequences, and optimized computational screening, are proposed to mitigate these challenges and ensure rapid response capabilities without compromising biosecurity. Second, a machine learning architecture for biological sequence modeling, “Lyra” is presented. Lyra is grounded in the biological principle of epistasis and leverages state space models (SSMs) combined with projected gated convolutions to efficiently capture both local and long-range sequence interactions. We demonstrate new mathematical theory to connect SSMs with the approximation of polynomial functions - key to predicting epistatic effects. This subquadratic architecture achieves state-of-the-art performance on diverse biological tasks, including protein fitness landscape prediction, RNA function prediction, and CRISPR guide design, while utilizing substantially fewer parameters and computational resources than existing foundation models like transformers. The thesis concludes by highlighting the synergistic potential of advanced machine learning and thoughtful policy to significantly improve pandemic preparedness.

In-or-Out: Creators’ Odyssey for Success

Thu, 01 May 2025 00:00:00 GMT

In-or-Out: Creators’ Odyssey for Success Li, Zelin The creator economy is flourishing, driven by shifts in advertising budgets and a surge in the supply of content creators. This has introduced a new challenge for firms: identifying which early-stage creators will grow to become stars. By identifying future stars, firms can choose who to invest their scarce resources in. They may also be able to purchase effective influence at a (proportionately) lower price than what they will pay once a creator becomes a star. Past research has shown that predicting which content will become viral is challenging. Instead, we focus on using content to predict which early-stage creators will grow their follower bases. We measure both the positioning of a creator’s early content and how the creator adjusts this positioning. We find that the initial position is not predictive of future success. However, subsequent adjustments in position are predictive, particularly if the creator’s initial follower base has grown consistently, rather than over a short period of rapid (viral) growth. Our insights inform the construction of predictive models that outperform baseline models in out-of-sample predictions of which creators will grow their followers the fastest.

Strategic Cooperation in Water Management: A Game-Theoretic Approach to Sustainable Infrastructure in Chilean Mining

Thu, 01 May 2025 00:00:00 GMT

Strategic Cooperation in Water Management: A Game-Theoretic Approach to Sustainable Infrastructure in Chilean Mining Moscoso Restovic, Rodrigo Y. Through a game-theoretic methodology this thesis examines collaborative approaches to managing water infrastructure within Chilean mining operations. The research examines cooperative stakeholder interactions to tackle water scarcity and growing demand in Chile's mining industry among mining firms, local residents and regulatory bodies. It utilizes game theory with a focus on cooperative games and bargaining models to develop a structured analytical framework for analyzing stakeholder dynamics including their incentives and cooperative opportunities. The thesis centers on creating a mathematical model that shows stakeholders as rational entities who seek to maximize their benefits while facing resource constraints and regulatory limitations. The implementation of cooperative game theory allows for detailed examination of coalition building processes along with resource sharing agreements and benefit allocation practices which helps to define stable cooperative possibilities. The primary findings show that mining companies achieve greater efficiency gains through water infrastructure collaboration than through separate individual investments. This thesis presents quantitative evidence that partnerships among mining projects generate significant financial savings and lead to better resource usage and positive environmental and social results. Sensitivity analyses identify that cooperative stability depends on several critical factors, including the asymmetries existing in the different mining projects, the sequence in which investment decisions are made, and the transfer price for water selling for those projects that prefer free rides. The final part of the thesis presents concrete suggestions for policymakers and industry leaders to develop cooperative frameworks through specific policy mechanisms and incentive systems that support long-term collaboration. The study advances existing academic knowledge by utilizing detailed game-theoretic approaches to address practical problems in sustainable mining practices. The findings reveal that strategic partnerships serve as fundamental tools for managing resources which can effectively tackle the urgent water scarcity challenges Chile faces.

Driving Manufacturing Best Practices Using Multimodal AI

Thu, 01 May 2025 00:00:00 GMT

Driving Manufacturing Best Practices Using Multimodal AI Zachary, Mark Multimodal artificial intelligence offers promising solutions for enhancing operational excellence in contract manufacturing, where small job shops typically operate with limited standardization and high process variability. This research develops a part similarity tool that integrates geometric, material, and scale information to improve quoting accuracy and engineering efficiency in high-mix, low-volume production environments. After examining the fragmented manufacturing landscape and reviewing current AI applications in manufacturing, the study introduces an approach based on Variational Autoencoders for encoding 3D geometry alongside material properties and dimensional scale information. The technical implementation addresses challenges of multimodal fusion, missing data handling, and computational efficiency, while a qualitative ablation study demonstrates how this comprehensive approach outperforms single-modal methods in manufacturing relevance. Engineers benefit from improved insights for manufacturing planning, while estimators achieve more consistent cost predictions using the multimodal system. Reinforcement learning with human feedback provides a mechanism for continuous refinement, creating a framework that bridges geometric similarity with manufacturing context and reduces subjectivity in critical business processes. The research contributes both theoretical insights into multimodal learning and practical implementation strategies for standardizing operations in contract manufacturing environments.

Made in Mexico: How Chinese Firms Navigate Nearshoring Amid Global Trade Disruptions

Thu, 01 May 2025 00:00:00 GMT

Made in Mexico: How Chinese Firms Navigate Nearshoring Amid Global Trade Disruptions Zeng, Bob This research explores the surge of Chinese manufacturing investments in Mexico as a strategic adaptation to recent global trade disruptions, specifically the U.S.–China trade tensions and the COVID-19 pandemic. By analyzing Chinese firms' motivations and strategies, the study highlights how they leverage Mexico’s strategic geographic proximity, favorable trade conditions under the USMCA, competitive labor market, and established industrial infrastructure to secure continued access to the North American market while minimizing tariff impacts and supply chain risks. Sector-specific analyses of the automotive, electronics, and renewable energy industries reveal distinct operational, regulatory, and cultural challenges encountered by these companies during their transition to Mexican production facilities. In addressing these challenges, Chinese firms have adopted strategies such as supply chain localization, rigorous adherence to North American regulatory frameworks, and effective cross-cultural management practices. Furthermore, the analysis situates this trend within the broader geopolitical context, emphasizing the role of evolving U.S. trade policies and proactive Mexican industrial initiatives in shaping the nearshoring landscape. The findings suggest that while Chinese investment in Mexico presents significant opportunities for industrial upgrading and enhanced bilateral cooperation, the longevity and effectiveness of these ventures depend on firms' strategic flexibility, deeper integration into local economies, and adept management of complex geopolitical and regulatory environments. By evaluating these elements, the research provides valuable insights into the drivers behind the increased Chinese presence in Mexico and the broader implications for global trade patterns, supply chain resilience, and regional economic integration.

Bridging the Gap: Strategies and Roles of Chinese Fintech Entrepreneurs and Startups in Sub-Sahara African Markets

Thu, 01 May 2025 00:00:00 GMT

Bridging the Gap: Strategies and Roles of Chinese Fintech Entrepreneurs and Startups in Sub-Sahara African Markets Zhu, Yuan This thesis examines the strategies and operational practices of Chinese fintech entrepreneurs in sub-Saharan African markets, with a focus on how they navigate regulatory fragmentation, localize business models, and build trust in low-infrastructure environments. Drawing on fieldwork and semi-structured interviews with founders, executives, and product leads from fifteen China-linked fintech firms across Nigeria, Kenya, and Francophone Africa, the study investigates how these actors engage with underdeveloped financial systems while adapting knowledge and models from China’s digital finance ecosystem. The research identifies several distinct approaches to market entry and adaptation, including platform integration, compliance-focused positioning, and informal ecosystem engagement. Findings suggest that these ventures do not simply export Chinese models but instead reconfigure them in response to local constraints in regulation, consumer trust, and institutional capacity. By analyzing firm-level strategies in diverse regulatory and market settings, this study contributes to broader discussions on transnational entrepreneurship, financial infrastructure development, and the evolving role of private actors in advancing digital inclusion across emerging economies.

A Data-Driven Work Center Assignment and Pricing Strategy for a Job Shop

Thu, 01 May 2025 00:00:00 GMT

A Data-Driven Work Center Assignment and Pricing Strategy for a Job Shop Carson, Alix Job shops with semi-autonomous work centers must understand their capacity utilization and financial state to maximize efficiency and profitability. Machine monitoring software allows managers to see the state of machines at any time and capture real-time capacity utilization. Job shops are positioned to maximize these work centers and must connect their manufacturing and operations strategy to the real-time shop data to maximize efficiency. This research is a case study in how a job shop can create a right-to-win strategy targeting jobs that are compatible and profitable for semiautonomous machines. ADDMAN Precision Baltimore (APBAL), a precision machine shop in the aerospace and defense industry, is facing labor constraints and underutilized work centers. This research aims to develop a structured quoting strategy and strategic pricing model to optimize job allocation between APBAL’s two semi-autonomous machining centers: the Makino Machining Complex 2 (MMC) and the Fanuc Robodrill. By integrating qualitative observations, historical job data, and machine utilization metrics, this study identifies inefficiencies in current job assignment practices. Key findings indicate that aligning work center assignments with projected profitability and capacity utilization can improve overall efficiency. A decision-making framework and pricing matrix are proposed to enhance job quoting accuracy, optimize machine usage, and increase APBAL’s competitiveness in securing high-volume contracts. The results offer a scalable framework for APBAL and its parent company, ADDMAN Engineering, to deploy across other machining facilities, ultimately improving operational performance and financial outcomes.

A Technoeconomic Model for Maritime Applications of Green Power Technologies

Thu, 01 May 2025 00:00:00 GMT

A Technoeconomic Model for Maritime Applications of Green Power Technologies Tuana, Daniel I. S. Growing societal and regulatory pressures are causing industries around the world to consider greener alternatives to conventional fossil fuel power technologies. As a result, power solution suppliers like CAT are facing strategic uncertainties: if, where, and when their core product markets will be disrupted by the novel adoption of alternative technologies. With the intention of helping to inform CAT’s future product and service strategy in conjunction with previous research related to powering mines and data centers, this thesis outlines the development of a code to estimate and compare the total cost of ownership of battery, hydrogen fuel cell, and nuclear power technologies to incumbent fossil fuel-driven systems in a variety of maritime scenarios including serving shoreside port electricity demand and on-water power demand across a diverse set of vessel segments. The code leverages first principles, empirical models, and researched assumptions to model the performance and costs of power systems in response to stochastically generated and deterministic power demand profiles over the useful lifetimes of the assets. For vessel applications, the code also estimates the volumes and masses of the alternative systems as a basis to judge their practicality. Hypothetical power systems for four archetypal ports and six vessel segments (across a range of power nodes) were studied to identify potential opportunities in and adjacent to the marine markets CAT currently serves. The outcomes of the study align with conventional intuition regarding the application of the technologies considered. Under certain conditions, the results support the technoeconomic case for the implementation of battery technology on short-haul vessels whose operations are predictable and would not be disrupted by shortened refueling/recharging intervals. Similarly, the results show that adoption of small modular nuclear reactors at ports and on large vessels with consistently large baseload power demand can provide economic advantages over incumbent fossil fuel technologies. The results of the simulations are sensitive to several technology-agnostic parameters including discount rates, fuel and electricity prices, demand growth rates, and other macro-economic conditions. In future, with ample case-specific data, the code developed for this thesis may provide convincing justification for the adoption of an alternative technology to serve the power demand of an individual port or vessel.

Discrete Event Simulation as a Predictor for Factory Traffic Management

Thu, 01 May 2025 00:00:00 GMT

Discrete Event Simulation as a Predictor for Factory Traffic Management Ramirez Echavarria, Esteban Manufacturing environments increasingly rely on automation and data-driven decision-making to optimize efficiency and production rates. This study explores the application of Discrete Event Simulation (DES) to model material flow and predict AGV (Automated Guided Vehicle), crane, and cart movements within a factory. The goal is to develop a digital twin that enables real-time decision-making, optimizes scheduling, and minimizes bottlenecks. To achieve this, we utilize SimPy, an open-source Python-based DES library, in conjunction with a custom-built API and React.js front-end interface. The study evaluates available DES software options and justifies the selection of SimPy based on flexibility, integration capabilities, and its suitability for modeling custom business rules. The solution is structured into modular components handling path planning, transporters, flows, stations, hot-cold starts, and utilities, ensuring adaptability to future improvements. A validation framework was established, utilizing historical data comparison and real-time validation to assess the simulation’s predictive accuracy. Over a 40-day testing period, the simulation achieved 89.6% accuracy and a sensitivity, or true positive rate (TPR), of 80.2%. The simulation provides a reliable first-pass scheduling tool that can be further refined with improved data collection. The findings indicate that while full automation of AGV deployment is not yet feasible, this study lays the foundation for future integration with the factory’s Vehicle Management System (VMS). Business implications include the potential for automated scheduling, enhanced material flow visibility, and optimization of capacity planning. Future work should focus on improving data accuracy, integrating live factory data streams, and refining algorithms for predictive scheduling.

From Strategy to Execution: An Optimization Approach to New Product Placement in the Apparel Industry

Thu, 01 May 2025 00:00:00 GMT

From Strategy to Execution: An Optimization Approach to New Product Placement in the Apparel Industry Netteberg, Sofie F. This thesis presents the development and implementation of a new product placement optimization model for a large global apparel and footwear company’s supply chain, aimed at maximizing network-wide profits while aligning with long-term strategic goals amidst demand volatility. The model leverages a mixed-integer linear programming approach, integrating probabilistic demand simulations to optimize the placement of new products within the company’s existing network of third-party partner company factories. Key elements of the model, including decision variables, price and cost coefficients, an objective function, and constraints that reflect operational realities and strategic priorities, are discussed in detail. Through analysis and results validation, this research demonstrates how data-driven optimization can improve network profitability and adherence to companies’ long-term strategic supply chain objectives and develop networks that are more profitable. The thesis then includes an exploration of historic demand variability at the host company, followed by a recommendation to integrate probabilistic forecasting in network planning to generate production networks more robust to volatility in consumer product demand. The findings contribute to advancing data-driven decision-making in supply chain management and offer actionable insights for future product placement strategies.

Policy Approaches and Entrepreneurial Responses in Strategic Industries: Comparing Innovation Ecosystems in China and the United States

Thu, 01 May 2025 00:00:00 GMT

Policy Approaches and Entrepreneurial Responses in Strategic Industries: Comparing Innovation Ecosystems in China and the United States Ni, Mengmeng This thesis investigates how government policy approaches shape regional entrepreneurial ecosystems and influence entrepreneurial strategy in strategic industries across China and the United States. Through comparative analysis of four region-industry pairs—Shanghai's semiconductor sector, Shenzhen's drone technology sector, Boston's biotechnology cluster, and New York's fintech ecosystem—the study examines the dynamic interplay between institutional design and entrepreneurial behavior. Drawing on Porter's Cluster Theory, Mazzucato's Entrepreneurial State concept, and the MIT REAP framework, the research develops a novel policy categorization encompassing four innovation governance tools: Cluster and Crisis Response Tools, Innovation Ecosystem Tools, Market-Shaping Tools, and Institutional Restructuring Tools. A qualitative case study methodology is employed, with in-depth firm-level analyses of Biren Technology in Shanghai and Moderna in Boston illustrating how entrepreneurs strategically respond to distinct institutional environments. The findings reveal four distinct models of innovation governance: Shanghai’s state-directed coordination, Shenzhen’s regulatory experimentation, Boston’s market-based orchestration, and New York’s regulation-centered oversight. Across contexts, entrepreneurs emerge as interpretive agents who actively leverage, adapt to, and at times reshape institutional conditions. This thesis contributes to the literature by offering comparative insights into the co-evolution of public policy and entrepreneurial strategy. It also provides practical implications for policymakers designing innovation ecosystems and for entrepreneurs navigating increasingly complex regulatory and technological landscapes.

Developing a Data-Driven Approach to Reducing Excess Inventory in a Multi-Echelon Supply Chain

Thu, 01 May 2025 00:00:00 GMT

Developing a Data-Driven Approach to Reducing Excess Inventory in a Multi-Echelon Supply Chain Gosen Cappellin, Carlos Daniel The medical technology company MedTechCo, specifically its Spine division, has deployed millions of implants in hospitals to meet demand. When inventory deployment and allocation are not managed appropriately to ensure that products are in the right place at the right time, excess inventory arises. Currently, MedTechCo Spine holds large amounts of excess inventory that are not utilized effectively. The objective of this research is to leverage a data-driven approach to define and reduce implant excess inventory at scale for MedTechCo’s Spine business unit in the United States. The research strategy used in this thesis begins with a root cause analysis to understand the causes of excess inventory. A robust data model was then developed to determine appropriate inventory levels by SKU, map all excess field inventory, and prioritize the most valuable excess SKUs. This data model was used to automate the company’s ERP system to repurpose excess inventory, limit unnecessary inventory deployments to the field, and eliminate redundant backorders. Finally, an impact analysis was performed to measure the potential excess inventory reduction in both dollar value and units. Time constraints limited the implementation of the recommendations during the research period. However, MedTechCo Spine agreed to incorporate the proposed recommendations into its ERP system and operational processes in mid-2025. These recommendations will help reduce implant excess field inventory, unlocking tied-up capital, creating flexibility in the supply chain to meet demand changes, and enabling additional investment in innovation.

AI and ML in Real Estate Underwriting: Transforming Financial Decision-Making and Operational Efficiency

Thu, 01 May 2025 00:00:00 GMT

AI and ML in Real Estate Underwriting: Transforming Financial Decision-Making and Operational Efficiency Jaklis, Cyril Real estate is the world's largest untapped market, at $650 trillion (Statista, 2023), yet technological innovation, particularly in financial underwriting, is underrepresented. Excel spreadsheets, broker-driven data collection, and expensive public database subscriptions are still used by most institutional players and family offices. These outdated approaches result in inefficiencies and higher operational expenses. Firms are now waiting for more innovative tools to improve their workflows and predict their Net Operating Income (NOI). Development and maintenance costs are often underestimated due to optimistic estimates and unplanned material or labor cost price escalations. This paper examines how to increase the accuracy of underwriting by examining the full underwriting process, identifying operational inefficiencies, and analyzing how new technologies like Artificial Intelligence (AI) and Machine Learning (ML) are currently being utilized to better value properties and reduce error margins. The analysis covers the entire underwriting process, from data sourcing, collection, structuring, and analysis. It also reviews the platforms and software tools utilized to connect these phases, from initial appraisal to investment memo and investment committee (IC) decision-making. The objective is to understand practical constraints, recognize opportunities for optimization, and explore where investors can strategically position themselves to leverage these technologies while also providing a forward-looking outlook on the changing function of AI/ML in the sector over the next decade.

Investigation Into Sources of Volatility in Sortation Center Processes to Improve Productivity and On-Time Delivery

Thu, 01 May 2025 00:00:00 GMT

Investigation Into Sources of Volatility in Sortation Center Processes to Improve Productivity and On-Time Delivery Fenstermacher, Andrew D. Target Corporation has expanded its Last Mile Delivery (TLMD) capabilities through an omni-channel, "stores-as-hubs" strategy, using stores as fulfillment centers for online orders. The Target Sortation Centers was developed to receive packages from stores in the region, sort, route and dispatch these packages each day to accomplish faster delivery for online orders. Designed to never hold inventory, the goal is to have every package received delivered that same day. This presents new operational challenges common for brick-and-mortar retailers that develop an omni-channel strategy. This thesis investigates core processes in Sortation Centers to identify sources of volatility and propose improvements that enhance productivity and on-time delivery while minimizing labor costs and incomplete volume. Many of the current processes in Target’s Sortation Centers are manual and unstandardized. Moreover, improving operations and piloting changes is challenging, especially during peak seasons. To address these challenges, this study employs discrete event simulation (DES) using SimPy, informed by current operational data and in-person observations, to model and analyze current processes. Key findings reveal that pre-sorting TLMD volume from other national carrier volume at the stores prior to linehaul pick up for same day packages decrease the overall completion times for the day’s volume by 5.8% and lowers incomplete volume probability by up to 85% under excess volume scenarios. These process changes enhance site resilience to demand volatility without significant capital investment. The research underscores the value of DES for testing process improvements virtually and highlights the need for network-level optimization across Target’s omnichannel supply chain. Recommendations include piloting floor loading and pre-sorting in select markets, alongside future exploration of performance standards, automation, and standardized processes to further mitigate volatility impacts.

Computer Vision for Cell Line Development

Thu, 01 May 2025 00:00:00 GMT

Computer Vision for Cell Line Development Albright, Jackson A. Anomalies in Cell Line Development prove to have significant impact on material and opportunity cost when screening for the Master Cell Bank that is used for all clinical drug development. Cell Line Development scientists spend hundreds of hours collectively identifying anomalies in fluorescent and brightfield imagery to ensure only high-performing cell clones are downselected for testing. The use of computer vision models alleviates this burden on scientists and better standardizes the selection process. Three techniques were tested for classifying anomalous and nominal fluorescent images: an autoencoder, an edge CNN and an RGB SVM. Examining performance through composite metrics such as F1 Score and MCC, the autoencoder (0.8744 and 0.8619, respectively) outperformed the edge CNN (0.8488 and 0.8257) and RGB SVM (0.8343 and 0.8252) for fluorescent anomaly classification. The high performance of the autoencoder came from training solely on anomalous images and using a percentile-based threshold to classify images on their reconstruction error. Data robustness proved to be an issue, with certain test datasets having worse performance due to inherent variability of images within both nominal and anomalous classes. Gathering and labeling more datasets for training and testing will allow models to learn from this variability and provide higher confidence in model performance for real-time screening applications. Adjusting the structure of the traditional autoencoder to that of a variational autoencoder will also help with learning the variability of images within classes, and improve performance on previously unseen data. Overall, the current iteration of the models proves to be beneficial for anomaly detection in Cell Line Development and demonstrates that some modifications to data sourcing and model architecture could see even better performance. These same techniques could be applied to similar biopharmaceutical applications provided care is taken to properly source clean and labeled image data and construct appropriate model architectures for the images' inherent features.

Sensor simulation for autonomous vehicles: Diffusion based image and depth generation for driving scenes

Thu, 01 May 2025 00:00:00 GMT

Sensor simulation for autonomous vehicles: Diffusion based image and depth generation for driving scenes Bieske, Linn Background: Autonomous vehicle (AV) testing requires extensive real-world data collection, which is costly and time-consuming. Existing simulation techniques struggle to generate high-fidelity sensor data, particularly for multimodal signals like RGB camera images, LiDAR depth maps or LiDAR point clouds. Recent advances in generative AI, specifically diffusion models, offer a solution for improving synthetic driving scene simulations. Objective: This thesis enhances diffusion-based generative models to: 1) Encode LiDAR depth data into a stable diffusion model’s latent space, 2) Generate simultaneously, consistently and with high fidelity eight RGB camera images, 2D LiDAR depth maps and 3D LiDAR point clouds for a full 360-degrees range, and 3) Evaluate the realism and consistency of the generated sensor data. Methods: A multimodal, multi-view latent stable diffusion model was trained to generate complete 360’ synthetic driving scenes and simulate camera and LiDAR sensor signals for autonomous vehicles. The generated scenes were evaluated for sensor alignment, realism, and depth accuracy. Results: The diffusion model produced realistic, spatially consistent camera and LiDAR sensor data, reducing reliance on real-world validation miles and lowering AV testing costs. To further improve the quality of the multimodal driving scene generation it is recommended to retrain the VAE on LiDAR data. Conclusion: This work advances AV simulation by extending stable diffusion models to multimodal sensor data. Future improvements should focus on real-time generation and expanding to additional sensor types or hardware setups for enhanced simulation fidelity.

Predictive Modelling of Customer Membership Purchases to Minimize Marketing Costs

Thu, 01 May 2025 00:00:00 GMT

Predictive Modelling of Customer Membership Purchases to Minimize Marketing Costs Liu, Ying This thesis develops and evaluates a series of predictive models to improve the efficiency of marketing resource allocation in the context of an outbound campaign for a premium membership product. The central objective is to identify customers most likely to respond positively to a membership offer, thereby minimizing outreach costs and maximizing return on investment. The study leverages a dataset from a large retail superstore that includes customer demographics, transactional behavior, and campaign response history. Data preprocessing involved the creation of engineered features such as age and tenure groupings and the transformation of categorical variables into factor types suitable for classification algorithms. Three modeling approaches were applied: classification with logistic regression, classification and regression trees (CART), and random forest. Logistic regression yielded strong predictive performance with an AUC of 0.851 and identified several statistically significant predictors, including spending on wine and meat products, recent purchase behavior, and tenure length. However, its primary limitation lies in its inability to accommodate cost asymmetries, as it lacks the capacity to incorporate a loss matrix which assigns different penalty to false positives and false negatives. The CART model addressed this limitation by introducing a customized loss matrix that reflects the asymmetric cost structure of marketing misclassifications—assigning a higher penalty to false negatives than to false positives. While this cost-sensitive structure aligned better with business objectives, the CART model achieved a moderate AUC of 0.767, reflecting limited classification accuracy and robustness. To overcome these limitations, a Random Forest model was implemented, combining the strengths of ensemble learning with cost-sensitive training. It achieved the highest AUC of 0.864 and allowed for the integration of a loss matrix during training. Feature importance analysis revealed that variables such as number of days since the last purchase, the amount spent on meat products, and a customer's enrollment length with the company were among the most influential predictors of customer response. The model not only improved classification performance but also supported strategic targeting through interpretable outputs. An economic evaluation demonstrated the practical value of the predictive model. Under a loss matrix where the cost of a false positive was set to $2 and a false negative to $10, the Random Forest model reduced total campaign costs by approximately 30% compared to a non-targeted approach. This cost savings translates into a meaningful economic impact, particularly when applied to large-scale campaigns. Overall, the findings support the use of Random Forest with a cost-sensitive design as a superior modeling framework in marketing applications. By aligning machine learning with real-world cost structures, this approach offers both statistical rigor and economic relevance for data-driven decision-making in customer acquisition strategies.

Data, Analytics, and Optimization for Production Planning

Thu, 01 May 2025 00:00:00 GMT

Data, Analytics, and Optimization for Production Planning Malinowski, Maxwell X. This thesis serves as a case study for the implementation of data analytics and optimization within a high mix low volume electronics production environment in the Aerospace and Defense industry. This case study demonstrates the benefits of data analysis for defining and quantifying operational bottlenecks and explores the implementation of an optimization model to better allocate resources for production planning. Results demonstrate the insights derived from using data and analytics in this environment, and further discussion explores what contributes to an effective implementation of an optimization model in a production setting.

Industrial Pollution and Firm Ownership Structure: Evidence from M&A

Thu, 01 May 2025 00:00:00 GMT

Industrial Pollution and Firm Ownership Structure: Evidence from M&A Zhang, Cindy This paper studies whether firm ownership structure influences pollutive activity. Using facility-level data from the Toxics Release Inventory, I employ a differences-in-differences (DiD) approach to compare toxic chemical release and pollution prevention activity between public and private firms' facilities by exploiting ownership changes. I compare facilities initially owned by private firms that were acquired by public firms and those that were acquired by private firms in the same year. My findings suggest that public acquirers significantly reduce toxic release activity relative to private acquirers. In the reverse case, I find that private acquirers decrease abatement, but pollution volume does not differ significantly. However, for later ownership changes in my sample, private acquirers increase toxic release volume and intensity significantly relative to public acquirers. Lastly, I explore how financial constraints and the local political environment moderate pollution activity. Debt-constrained public acquirers show no significant difference in pollution activity from private acquirers. In Democrat-leaning counties, public acquirers reduce toxic releases more than private acquirers, but in Republican-leaning counties, the differences are less pronounced. Overall, my findings suggest that public firms have decreased toxic release activity over time, but the declines have been offset by increases from private firms.

Debt Complexity and Equity Behavior

Thu, 01 May 2025 00:00:00 GMT

Debt Complexity and Equity Behavior Li, Jack I examine how the complexity of firm debt affects the incorporation of news into equity prices. As residual claimants to firm cash flows, equity investors must be able to value all outstanding debt contracts, suggesting that complex debt can interfere with their ability to process news effectively. Using a model in which debt complexity causes a subset of investors to initially underweight news precision, I derive three predictions for the equity behavior of debt-complex firms around news events: (1) they exhibit greater post-announcement drift, (2) they show elevated trading volume both on announcement day and in the post-announcement period, and (3) their return volatility decreases on announcement day but increases during the post-announcement period. These predictions are supported by empirical evidence in the context of earnings announcements, suggesting that debt complexity introduces meaningful frictions in how news is incorporated into equity markets.

Mapping Wildness: Simulating Post-Extraction Wildland Regeneration

Wed, 01 May 2024 00:00:00 GMT

Mapping Wildness: Simulating Post-Extraction Wildland Regeneration Griggs, Crystal Ling This thesis introduces a novel approach to wildlife habitat classification for ecological regeneration. It is focused by the extreme environmental degradation of mountaintop removal (MTR) in the Appalachian Mountains, a violent coal extraction process that has significantly altered the landscape of this ecologically sensitive region. By integrating remote sensing and Geographic Information Systems (GIS) with machine learning, this research aims to develop a method that transcends traditional human egocentric landscape assessments, advocating for a model that foregrounds the habitats and needs of critically endangered species by simulating landscape regeneration and assessing topographical alterations in terms of how design decisions impact wildlife. Central to this study is the concept of Umwelt, the subjective experiences of nonhuman species, including how their spatial perception and spectrum are used to discern details within their environment. Umwelt broadens traditional spatial understanding by emphasizing that each species experiences the world through its sensory filters, which shape its interactions within their habitat. This understanding guides the research’s approach to approximating the Umwelt of the Cerulean Warbler (Setophaga cerulea), a surrogate species in this work, which has faced steep declines due to habitat loss in Appalachia. Through the development of a habitat suitability model that utilizes advanced computational tools and multispectral imagery, the thesis endeavors to offer a new perspective on environmental planning and conservation efforts - a computational approach to near-approximations of Umwelt. The methodological framework seeks not only to classify post-extraction landscapes for their potential in supporting wildlife but also to inform design and land use decisions that are sensitive to the temporal and complex processes of natural habitat regeneration. By challenging the prevailing paradigms of landscape restoration, which often lack consideration for the intricacies of wildland dynamics such as the multitudes of species interactions and interdependencies, this research proposes a new methodology that empowers wildlife to guide the ecological recovery process. The findings underscore the potential of applied GIS and machine learning in environmental advocacy, setting a precedent for future research and practice aimed at the regeneration of ecosystems that considers the ecological realities of all species involved.

Hydrogen Adoption Dynamics: A Flexible Modeling Framework for U.S. Industrial Applications

Thu, 01 May 2025 00:00:00 GMT

Hydrogen Adoption Dynamics: A Flexible Modeling Framework for U.S. Industrial Applications Ray, Jennifer As climate change concerns drive the need for decarbonization, hydrogen stands as a potential tool to help reduce emissions across the United States industrial and energy sectors. This thesis develops a flexible modeling framework for hydrogen adoption across multiple industrial applications, designed specifically to support strategic investment decision-making in an evolving market. The tool analyzes six major industries – steel, chemicals, energy storage, biofuels, vehicles, and natural gas– through two metrics: potential hydrogen consumption and threshold prices for economic viability. The framework applies scenario analysis to examine how government policy and technological advancement influence potential market trajectories. Analysis reveals significant sensitivity to input assumptions. Even small variations in the assumed initial hydrogen production cost can result in significantly different adoption timelines. In scenarios where initial hydrogen production costs are $5/kg, widespread adoption requires maximum policy support and technological progress. However, reducing the initial cost by just $1 to $4/kg makes broader adoption feasible with less reliance on government intervention. Light-duty fuel cell electric vehicle penetration rate and steel industry growth rate emerge as the most sensitive parameters affecting overall hydrogen demand, followed by biofuel blending rate and hydrogen injection percentage into natural gas infrastructure. The vehicles industry is identified as a first mover in widespread hydrogen adoption, followed by steelmaking and methanol production. Hydrogen adoption for natural gas blending, methanol for export, and methanol-to-gasoline applications occur later due to their lower threshold price for economic viability. Under optimal conditions with strong government support and significant technological advancements, total hydrogen demand could reach 48.8 million metric tons by 2050, approximately a sevenfold increase from scenarios with minimal support. The tool’s value lies not in projecting a definitive, single-point forecast, but in providing a flexible framework that helps stakeholders navigate market uncertainties as the decarbonization landscape evolves. Future research should integrate supply-side dynamics, infrastructure requirements, and geographic variability to enhance projection accuracy.

Towards Green Aluminum

Thu, 01 May 2025 00:00:00 GMT

Towards Green Aluminum Schurr, Kevin Aluminum is an important metal to facilitate the energy transition. Its high strength to weight ratio and easy recyclability make it a useful material in many industries from automobiles to food packaging. However, the aluminum smelting process accounts for 2% of all global greenhouse gas emissions due to both the high amount of power needed to facilitate the electrolysis reaction and to the consumption of carbon anodes in the process. As regulatory changes in Europe raise the monetary cost of emitting carbon, smelters are investigating new technologies to integrate into their operations to cut Scope 1 and 2 emissions. Two such technologies are carbon capture systems to abate process emissions and small modular nuclear reactors to reduce emissions incurred during electric power generation. This work explores the technical and economic feasibility of leveraging these systems at Aluminum of Europe, a primary aluminum smelter subject to these changing European regulations. Results suggest that while these technologies have not been specifically adapted for aluminum production yet, they can play an important role in reducing the overall emissions from the smelting process under specific economic conditions. However, the analysis indicates that, at present, significant subsidies are required for such projects to be financially viable.

Corporate Transparency and Cybersecurity Risks

Thu, 01 May 2025 00:00:00 GMT

Corporate Transparency and Cybersecurity Risks Kim, David Sunghyo I study whether disclosure mandates alter the equilibrium of cyberattacks by unintentionally informing cybercriminals. The California Consumer Privacy Act (CCPA) requires companies to disclose their personal information collection practices to consumers, inadvertently informing cybercriminals about the potential benefits of breaching each firm. Using a difference-in-differences design, I find that firms disclosing the collection of valuable personal data face an increased probability of data breaches. These firms also strengthen their cyberdefenses both in terms of cybersecurity software and cybersecurity specialists. Firms trade off cybersecurity costs against the risk of data breaches, with the increase in breach probabilities more pronounced among firms that invest less in cybersecurity. Finally, I find that firms adjust their data collection policies as additional defense strategies. Overall, this study highlights the trade-off between transparency and cybersecurity risks in today’s digital economy.

Fully Connected Digital Ecosystems within Hospitals – AI/ML Solutions for Improved Patient Care

Thu, 01 May 2025 00:00:00 GMT

Fully Connected Digital Ecosystems within Hospitals – AI/ML Solutions for Improved Patient Care Dugan, Andrew D. Cardiogenic shock (CS) in the context of acute myocardial infarction (AMI) remains a significant challenge in critical care, with high mortality rates despite the availability of advanced mechanical circulatory support (MCS) devices like the Impella pump. However, adoption of these devices in clinical practice remains limited. This thesis explores two complementary strategies to address these challenges: developing machine learning (ML) models to predict shock severity and assessing the feasibility of integrating hospital Electronic Medical Record (EMR) data into Abiomed’s digital ecosystem to support standardized shock care. In the first phase, ML models were trained on multiple clinical datasets to predict Society for Cardiovascular Angiography and Interventions (SCAI) shock stages based on patient data. While these models demonstrated strong predictive performance, feature analysis revealed that SCAI stages often reflect physician treatment decisions rather than purely patient physiology. This raises concerns about their utility as real-time clinical decision tools and suggests that ML applications may be better suited to prompting early data collection and intervention before severe shock develops. The second phase evaluated the feasibility of EMR integration to support the broader adoption of standardized shock protocols. After considering regulatory, operational, and technical factors, third- party data aggregation emerged as the most practical path forward. Integrating EMR data could improve outcome tracking, support protocol adoption, and strengthen partnerships between Abiomed and hospitals, creating a foundation for more consistent and proactive shock management. Together, these findings highlight the need for predictive tools that guide early clinical action and infrastructure that supports seamless data integration. By advancing both, Abiomed can expand its role in cardiogenic shock care, improve patient outcomes, and lead the evolution of data-driven, standardized treatment strategies.

Strategic Recommendations for Legacy Automakers in the Evolving Automotive Landscape

Thu, 01 May 2025 00:00:00 GMT

Strategic Recommendations for Legacy Automakers in the Evolving Automotive Landscape Tike, Gauri The automotive industry is undergoing a transformative shift with technological advancements in many areas such as Electric cars, Autonomous vehicles, Software Defined Vehicles, and decarbonization of mobility. Alternate means of transportation are also becoming available and sometimes even the cost is even lower than owning a car. The best way to get from point A to point B might not be the car in some of the cities. It might involve heterogenous modes of public transportation, using a bike, using ride hailing service or using a car for different portions of the route. Despite the concerns about the environment, we are still seeing an increase in global car ownership trends. These changing times pose challenges to legacy automakers. While they are experts in traditional car manufacturing, modern cars not only require traditional mechanical and electrical skills but also need deep expertise in developing software for these cars. With the growing EV adoption, we are seeing Chinese EV automakers are capturing market share quickly. What is the future of mobility with all these developments? What do traditional automakers need to do in this era to remain successful? In this report we will examine key trends in mobility: Global electric vehicles (EVs) adoption, software-defined vehicles (SDVs), autonomous vehicles (AVs), environmental implications. Based on this research we will propose strategic recommendations for traditional automakers in order to continue their success over the next decade and beyond.

Navigating Fintech Innovations: Strategic Insights from the United States and India

Thu, 01 May 2025 00:00:00 GMT

Navigating Fintech Innovations: Strategic Insights from the United States and India Shanbhag, Rishabh Ganesh This thesis examines how fintech ventures are reshaping financial services through new technologies and strategic choices tailored to different markets. It first looks at key innovations: digital payments, digital wealth management, and open banking, and how they have transformed everyday financial activities. The research then compares how fintech companies operate in the US and India by analyzing how market conditions, government initiatives, regulations, and consumer behaviors shape adoption. Finally, through case studies of Robinhood (US), Revolut (Global), and Paytm (India), the thesis examines how fintech firms navigate the choice between competing with traditional players and collaborating with them to scale under different market scenarios. Together, these insights aim to help entrepreneurs, investors and policymakers understand how strategy and technology come together in the fintech industry.

Forming the Future: A Digital Approach to Simulating Thermoplastic Manufacturing

Thu, 01 May 2025 00:00:00 GMT

Forming the Future: A Digital Approach to Simulating Thermoplastic Manufacturing Harkavy, Rachael This thesis develops a digital framework for simulating and validating thermoplastic composite manufacturing processes, focusing on reducing the time associated with new product development. Using Finite Element Analysis (FEA) software (SimSof) and high-precision 3D scanning tools (ScanSof), the research introduces a geometric similarity metric to quantify deviations between simulated and real-world parts. By aligning simulations with production data, the study aims to replace costly physical trials with reliable digital models, accelerating customer onboarding and improving manufacturing efficiency. Key contributions include establishing a systematic pipeline for integrating simulation tools into Oribi Composites’ workflow, defining critical parameters such as laminate width, material card accuracy, and mesh size, and validating their impact on simulation accuracy. Results demonstrate that accurate material modeling and parameter selection significantly enhance digital twin accuracy, while mesh size has minimal influence, allowing for computational cost savings. The research also highlights challenges in replicating real-world conditions digitally, including inconsistent material cards, and limited control over pressure profiles. Despite these limitations, the study proves that simulations can reliably predict manufacturable designs within customer tolerances, reducing reliance on physical iterations.

Toward a Sustainable and Scalable Ecosystem: Breaking the Cycle of Intergenerational Poverty for Single Mothers in Japan with Private Sector Engagement

Thu, 01 May 2025 00:00:00 GMT

Toward a Sustainable and Scalable Ecosystem: Breaking the Cycle of Intergenerational Poverty for Single Mothers in Japan with Private Sector Engagement Imaeda, Hiroko Despite Japan’s reputation as an economically advanced nation, it faces one of the highest relative poverty rates among OECD countries, with nearly half of all single-mother households living below the poverty line. This thesis examines why poverty among single mothers persists despite a formal support ecosystem and proposes a systemic redesign grounded in life-stage-aligned, user-centered principles. Drawing on historical-institutional analysis, organizational theory, fieldwork interviews, and auto-ethnographic insights, the study identifies deeply embedded barriers that reinforce fragmented, crisis-oriented support systems misaligned with real-life trajectories. In response, it introduces the "Single Mother Journey" framework, reframing single mothers not as a static category but as a dynamic population with distinct, evolving needs. Through this lens, the thesis exposes critical gaps in preventive support, labor market misalignment, and information accessibility. Building on these findings, it proposes a future-ready support ecosystem, positioning corporations, local municipalities, NPOs, and education institutions as collaborative actors. It presents mumtec, a conceptual digital platform designed to consolidate fragmented services, personalize interventions by life stage, predict crisis points, and generate adaptive policy feedback. The thesis moves beyond surface-level critique by connecting institutional analysis with practical system design to offer a scalable framework for inclusive innovation. Listening to the silent voices of single mothers navigating precarity is an ethical imperative and a strategic necessity for sustainable, resilient societies.

Optimizing automotive production scheduling to reduce finished vehicle inventory

Thu, 01 May 2025 00:00:00 GMT

Optimizing automotive production scheduling to reduce finished vehicle inventory Johnson, Christopher This thesis addresses inefficiencies in automotive finished vehicle inventory management arising from misalignment between production scheduling and outbound logistics. Traditional production planning prioritizes manufacturing efficiency, causing significant inventory accumulation as vehicles await completion of full shipment loads. This research proposes an Integrated Production and Outbound Distribution Scheduling approach, introducing an optimization step within existing production scheduling workflows to align production sequences for expedited load formation. Back-testing on two automotive assembly lines over 82 weeks reveals a mean inventory reduction potential of 63–65%, with variability influenced by production volumes and vehicle configurations. A proof-of-concept implementation confirms the practical feasibility of optimized schedules, reducing inventory holding times by 33% without disrupting manufacturing operations. Computational performance analysis demonstrates good scalability for instances with fewer than 600 vehicles, though larger scenarios still yield meaningful inventory reductions. This work highlights substantial opportunities for automotive original equipment manufacturers to enhance efficiency by integrating outbound logistics into production scheduling.

Transforming Unstructured Data into Actionable Insights: A Use Case of Generative AI in Operational Technology Problem Management

Thu, 01 May 2025 00:00:00 GMT

Transforming Unstructured Data into Actionable Insights: A Use Case of Generative AI in Operational Technology Problem Management Gallardo Moncayo, Gabriel A. The increasing availability and reduced cost of Generative AI applications for the general public have motivated organizations across all industries to implement AI-based solutions in their daily operations. Still, they struggle to determine the capabilities and limitations of this technology when implementing it in their specific context. This thesis addresses these challenges through a practical case study: deploying a text-based Generative AI system (using Large Language Models - LLMs) for automated downtime event characterization within a global industrial operational technology (OT) setting by transforming unstructured problem management reports into structured, actionable business insights. The developed software system contains a data pre-processing stage, followed by four LLM-based tasks (LLM-extraction, LLM-autoclassification, multi-aspect multi-level LLM-classification, and LLM-accuracy). We wrap everything in a well-structured and easy-to-understand evaluation framework that ensures the system’s output is format-reliable, accurate, and consistent. Through simple prompt engineering techniques and continuous failure modes analysis, we achieve high accuracy (>89%) and consistency (>79%) for downtime events characterization at 1% of the current cost. In the end, we prove that it is possible to implement an AI-based solution within current operational processes while properly communicating its capabilities and limitations and adapting its usage to the most added value purpose.

Optimizing Raw Wire Inventory Management: A Data-Driven Approach to Demand Forecasting and Supply Chain Decision Support

Thu, 01 May 2025 00:00:00 GMT

Optimizing Raw Wire Inventory Management: A Data-Driven Approach to Demand Forecasting and Supply Chain Decision Support Gebner, Adam R. This thesis investigates methods to improve demand forecasting and inventory management for raw wire. Challenges such as supply chain disruptions from the COVID-19 pandemic, operational variability, and loss of expertise exposed vulnerabilities in the existing manufacturing system, leading to shortages and inefficiencies. By leveraging extensive production data, this research develops and evaluates tools to mitigate these issues while aiming for a 100% service rate. The project leveraged extensive production data to predict future wire requirements, optimize inventory, and achieve a 100% service rate. Key contributions include: 1. A data-driven demand simulation model, reducing forecast error and surpassing baseline methods 2. Quantification of waste distributions and variability in wire consumption 3. An inventory simulation framework for policy evaluation and shortage mitigation 4. Clustering analysis to classify demand patterns and identify key wire categories 5. A decision support tool supporting real-time visibility into inventory levels and risks The models and tools developed through this project provide enhanced capabilities to predict future wire requirements and manage inventory more effectively through continued development. Though the initial results indicate potential business value, areas for future work include incorporating additional data sources, exploring advanced machine learning techniques, and conducting longer-term pilot studies to quantify business impact. This project demonstrates the value of leveraging data analytics and simulation modeling to enhance supply chain decision-making in complex manufacturing environments.

Economies of Space: Developing a Lean Manufacturing Framework for Work Center Floorspace Reduction

Thu, 01 May 2025 00:00:00 GMT

Economies of Space: Developing a Lean Manufacturing Framework for Work Center Floorspace Reduction Gerbino, Jacob This thesis aims to develop a lean manufacturing framework with the goal of optimizing the use of floorspace in Boeing's Interiors Responsibility Center South Carolina (IRCSC). The primary goal is to eliminate wasted floorspace while increasing production capacity and efficiency. The motivation behind this project stems from the need to address the fully allocated production floorspace at IRCSC and the pressing requirement to add new product lines without expanding the facility's physical footprint. Additionally, the project seeks to prepare IRCSC for possible increases in production rates for the 787 Dreamliner Program, necessitating a redesign of work centers to support higher output levels while enhancing efficiency and reducing costs. The project employs the DMAIC (Define, Measure, Analyze, Improve, Control) methodology and lean tools such as spaghetti diagramming and value stream mapping to treat "Misused Space" as an additional form of waste, alongside the traditional forms of lean waste. The framework was applied to a sample interior product work center to test its effectiveness. The study involved mapping the current layout, observing technician travel, conducting time studies, and analyzing value stream maps. The methodology facilitated the creation of a new floorplan and scheduling system that consolidates cure times and balances workloads between work cells. Discrete event simulation was used to validate the proposed changes, ensuring they would achieve the desired improvements. The results of the study revealed inefficiencies in the current layout and scheduling practices of the work center. The proposed changes demonstrated a potential 25% reduction in floorspace and a 55% decrease in product throughput time. The new scheduling and work allocation strategy reduced product throughput time from nine days to four, and the new layout reduced worker travel distances by as much as 50% in some work cells. The lean manufacturing principles and scheduling optimizations discussed in this thesis should be applied to other work centers within IRCSC. Future research should explore advanced methodologies and tools to handle the complexities of more interconnected work centers.

The Impact of AI Integration in Healthcare: Exploring Regulatory, Cultural, and Strategic Barriers

Thu, 01 May 2025 00:00:00 GMT

The Impact of AI Integration in Healthcare: Exploring Regulatory, Cultural, and Strategic Barriers Venkatanarayanan, Sriya This thesis investigates the barriers and enablers to predictive AI adoption in healthcare through a thematic synthesis of 13 academic articles and real-world case studies published over the last five years. Barriers were categorized into three domains: regulatory, cultural, and strategic. These included challenges such as fragmented regulation, clinician skepticism, data quality limitations, and poor alignment with clinical workflows. Cross-cutting patterns, stakeholder tensions, and recurring meta-themes revealed that these barriers are deeply interconnected. Drawing from over 200 individual findings, an actionable visual framework was developed to guide responsible and sustainable predictive AI integration. The proposed model, consisting of an internal “Pyramid” of enablers and an external “Circular Loop” of ecosystem conditions, provides a practical structure for aligning governance, engagement, and workflow with ongoing commitments to equity, collaboration, safety, and transparency.

Generative AI in Private Equity for Accumulative Advantage

Thu, 01 May 2025 00:00:00 GMT

Generative AI in Private Equity for Accumulative Advantage Mahajan, Bonny This research explores the use of Generative AI (Gen AI) for achieving accumulative gains across various business and technical functions within commercial enterprises under private equity firms. While based on applied experiments in a private equity-owned, resource constrained portfolio company, many of the findings presented here may apply in other types of organizations.Through this study, we conduct case studies across key departments such as customer service, purchasing, engineering, employee management, and marketing. For each use case, we delve into the utilization of custom-built or publicly available Gen AI-based tools, aiming to understand the unique considerations and challenges that may arise when implementing Gen AI solutions in industries like manufacturing, which have traditionally been underserved by the tech sector.Through this research, we identify the critical role of humans in the loop, emphasizing the importance of UI/UX design, domain expertise, and local culture in the successful adoption and acceptance of Gen AI tools designed to enhance workforce efficiency in portfolio companies. This study also aims to illustrate how investing in Gen AI technologies is ultimately an investment in a company’s most valuable resource—its employees. By equipping employees with innovative tools, the organization not only improves productivity and job satisfaction but also fosters a culture of continuous improvement and adaptability. This research highlights the transformative potential of Gen AI in reshaping traditional business processes and driving sustainable growth in different functions of organizations.

Standard Work for High Mix Low Volume Manufacturing

Thu, 01 May 2025 00:00:00 GMT

Standard Work for High Mix Low Volume Manufacturing McNulty, Will This thesis examines the challenges of developing standard work at scale in a high-mix low-volume (HMLV) manufacturing environment. The research is conducted at Re:Build Composite Resources, a thermoset composites (TSC) manufacturer. In the context of the company, impending growth demands more skilled laminators and the manual, complex nature of TSC lamination exposes the need for improved and documented standard procedures. By documenting existing processes through operator shadowing, time studies, and quality data analysis a “best-known” standard was created for the production steps of a subset of parts. Two pilot parts—one focused on cutting scrap rates, the other on boosting throughput—demonstrated how standard work instructions and a standard work schedule designed for one-piece flow significantly reduced errors and production variability. The thesis also explores the effectiveness and limitations of using computer vision as a tool to automate work instruction and time study data set generation. Beyond the immediate improvements in quality, efficiency, and new operator onboarding, the project’s scalable framework lays out a roadmap for broader adoption of standard work in fast-growing HMLV operations. By focusing first on parts that yield the most significant gains — either due to high volume or high unit cost — organizations can maximize returns on continuous-improvement efforts while not overburdening their engineering staff with excess analysis and documentation.

Evaluating The Feasibility of Electrified Process Heating for Drug Substance Manufacturing

Thu, 01 May 2025 00:00:00 GMT

Evaluating The Feasibility of Electrified Process Heating for Drug Substance Manufacturing Bhirgoo, Priya Darshini The pharmaceutical industry relies on high-temperature fluids such as pure steam to support critical operations including equipment cleaning and sterilization and on hot Water-For-Injection (WFI) as a key ingredient for drug substance manufacturing. These high-temperature process-driven heat demands are fulfilled through fossil fuel-based heating which contributes significantly to Scope 1 carbon emissions. Recognizing the link between environmental stressors and human health, Amgen has committed to achieving carbon neutrality by 2027. This thesis explores the feasibility and implications of transitioning from fossil fuel-based process heating to a fully electric system at one of Amgen’s drug substance manufacturing sites. Amgen’s existing fossil fuel-based steam system was analyzed through site visits, engineering reviews, and stakeholder engagements to quantify capital and operating costs, energy usage, and carbon emissions. A fully electric alternative was designed by researching commercial technologies and collaborating with suppliers as well as internal stakeholders. The analysis found that while the capital investment required for electrification is comparable to that of traditional steam systems, the operating costs for an electric system are significantly higher, driven by the higher price of electricity relative to natural gas. From a sustainability perspective, electrification eliminates on-site Scope 1 carbon emissions but shifts emissions to Scope 2, making the environmental benefit dependent on the carbon intensity of the local electricity grid. As grids transition to renewable energy sources, the potential for long-term emissions reductions strengthens. Future work should focus on evaluating the costs of necessary electrical infrastructure upgrades and identifying regions with lower-carbon, lower-cost electricity grids where electrified systems could be more readily implemented.

Partnerships as Retention Levers: A Study of Credit Card–Entertainment Collaborations

Thu, 01 May 2025 00:00:00 GMT

Partnerships as Retention Levers: A Study of Credit Card–Entertainment Collaborations Tchelikidi, Cloe In mature, competitive sectors such as financial services and media and entertainment, customer loyalty is increasingly difficult to sustain. This thesis explores the emergence of cross-industry partnerships, specifically between credit card issuers and digital entertainment platforms, as a strategic response to rising churn and declining differentiation. Through a case study of the American Express Digital Entertainment Credit, the research examines how lifestyle-aligned benefits can foster deeper behavioral engagement, reduce switching, and enhance customer lifetime value. The thesis situates these partnerships within the broader evolution of loyalty strategies, marked by hyper-personalization, subscription fatigue, and platform convergence. Findings suggest that flexible, recurring rewards embedded in consumers’ daily routines offer a path to durable retention, especially among younger, digital-native cohorts. The study concludes that such partnerships are not peripheral marketing tools but increasingly core to competitive strategy in commoditized markets.

Exploring the Dynamics of Regulatory Compliance, Cost Management, and Competition in the Pharmaceutical Industry

Thu, 01 May 2025 00:00:00 GMT

Exploring the Dynamics of Regulatory Compliance, Cost Management, and Competition in the Pharmaceutical Industry Wu, Lanchen This paper explores how financial pressures, regulatory enforcement, and market dynamics interact to shape pharmaceutical manufacturing quality and drug supply stability. Using a causal loop diagram (CLD), it examines how cost-cutting behavior affects control and validation capabilities, interacts with regulatory agency oversight, and contributes to recurring drug shortages. The analysis highlights how competition drive companies to operate at or near the minimum regulatory requirements, gradually eroding quality systems. Because of the nature of medical products, the quality of a drug cannot be directly assessed by individual users, distributors, or payers, making it necessary for government agencies like the FDA to rely on internal manufacturing data to ensure all drugs meet a minimum standard of quality. Regulatory oversight serves as a safeguard rather than a tool for guiding business decisions. However, its effectiveness is constrained by the frequency of inspections, the capacity of auditors, and limited resources—especially when government budgets are stretched and other priorities take precedence. The paper also discusses how manufacturers may avoid detection by strategically presenting information during inspections, making it harder for auditors to spot issues and allowing weakened controls to persist. Over time, these dynamics reinforce one another, creating a self-sustaining cycle in which cost pressures lead to a minimal compliance, quality issues, and regulatory responses that increase costs further. As the number of manufacturers shrinks due to market consolidation, supply disruptions become more severe when failures occur. Regulatory discretion—intended to avoid immediate shortages—can unintentionally reduce incentives for long-term quality investment, further weakening the system’s resilience. To address these issues, the paper proposes structural changes, including financial accountability for payers during shortages, tighter regulatory focus on process reliability, and linking regulatory flexibility to quality improvement obligations. These approaches aim to create balancing mechanisms that reduce cost-driven deterioration of quality and promote a more stable pharmaceutical supply chain.

Ant Group’s Transformative Impact on China’s Financial Industry

Thu, 01 May 2025 00:00:00 GMT

Ant Group’s Transformative Impact on China’s Financial Industry Pan, Kathryn Ant Group, China’s leading digital finance company, has fundamentally transformed the nation’s financial industry through groundbreaking innovations in digital payments, micro-lending, wealth management, and investment advisory. This paper explores the company’s role in reshaping China’s financial ecosystem, analyzing its impact on traditional banking institutions, regulatory policies, and consumer behavior. Utilizing analytical frameworks such as Porter’s Five Forces, PEST analysis, and SWOT analysis, this study provides a comprehensive assessment of the external and internal factors influencing Ant Group’s development and competitive positioning. This research highlights Ant Group’s key financial innovations, including its online transaction platform, offline payment services, online credit solutions, digital fund distribution channels, and AI-driven investment advisory. By leveraging advanced technologies such as artificial intelligence, blockchain, and big data analytics, Ant Group has enhanced service efficiency, expanded accessibility, and strengthened risk management capabilities. These innovations have significantly advanced financial inclusion, extending financial services to previously underserved populations. However, Ant Group’s rapid growth has also intensified regulatory scrutiny, prompting major restructuring efforts and adjustments to its business model. This paper employs three major analytical frameworks: PEST analysis, Porter’s Five Forces, and SWOT analysis. The PEST analysis examines the political, economic, social, and technological factors shaping Ant Group’s trajectory, highlighting the impact of evolving government policies and macroeconomic conditions on its operations. Meanwhile, Porter’s Five Forces framework assesses the competitive dynamics within China’s financial sector, identifying key market pressures such as rising competitions and regulatory constraints. Finally, the SWOT analysis evaluates Ant Group’s internal strengths and weaknesses, as well as external opportunities and threats, offering a comprehensive perspective on the company’s strategic positioning. Drawing from these analyses, the paper offers strategic recommendations to ensure Ant Group’s sustained growth and resilience in an increasingly complex financial environment. These recommendations include strengthening regulatory compliance, fostering strategic alliances with both domestic and international partners, and further leveraging technological advancements to expand its service offerings. Additionally, the study explores potential global expansion strategies, considering how Ant Group can adapt its innovative financial solutions to international markets while navigating diverse regulatory landscapes. By examining Ant Group’s evolution and the broader implications of its digital finance model, this study contributes to a deeper understanding of fintech’s disruptive power in China’s financial sector. The findings provide valuable insights for industry leaders, policymakers, and scholars interested in the intersection of financial technology, regulation, and strategic business management. As digital finance continues to evolve, Ant Group’s trajectory serves as a critical case study in balancing innovation, regulation, and market competition within a rapidly shifting financial landscape.

Process Optimization and Proactive Quality Control to Increase Investment Casting Throughput

Thu, 01 May 2025 00:00:00 GMT

Process Optimization and Proactive Quality Control to Increase Investment Casting Throughput Sircar, Julia Sarita Blue Origin is an aerospace company with ambitious throughput goals in response to increased commercial space exploration. Pressure to increase throughput is especially apparent within its BE-4 engine business, as the engines support Blue Origin and its customers. Blue Castings is one of the primary in-house manufacturing plants that supports BE-4 production; the plant manufactures rocket engine components through a process called investment casting. Investment casting, by nature, is a complex process involving long rework times, high incidence of defects, and significant process variability. These characteristics contribute to the discrepancies between Blue Origin’s target BE-4 production rate, the production rate feasible at Blue Castings, and its actual delivery rate. This thesis explores how defect management and prevention techniques can improve throughput at Blue Castings and reduce the number of Blue Origin’s schedule delays attributable to Blue Castings. The research began with a baseline investigation and analysis of Blue Castings’ actual and best-case throughput rates compared to its goal. Two gaps were identified: 1) a gap between actual and feasible throughput, and 2) a gap between feasible and target throughput. The analyses highlight the need for better process and quality management to close both gaps. Through a mixed-method approach, the researcher explored and piloted process and data improvements to understand their impact on throughput. This included qualitative and quantitative data collection through on-site interviews, random sampling of defect data, and queries from the manufacturing execution system. With this data, the researcher investigated how machine learning can predict rework severity and support defect prevention. A case study on a selected part number demonstrated the potential to improve throughput by reducing unnecessary rework. By aligning stock-on surface criteria to downstream machining requirements, average rework loops were reduced from thrice the industry benchmark to below the benchmark. This increased capacity at the rework work center and improved the overall delivery of this part. The research also demonstrated how a cross-functional collaboration to formalize producibility lessons reduces the creation of defects, promotes systematic knowledge-sharing, and accelerates improvements similar to the stock-on surface case study. In parallel, this research evaluated how Blue Castings could improve defect documentation and tracking without causing significant additional effort for operators. The researcher’s findings highlight the limitations of handwritten weld maps and inconsistent data capture practices on effectively preventing defects. Digitization of defect tracking is recommended to enable consistent defect data collection and improved root cause and trend analyses. As data quality improves, applying classification ML models for predictive analytics can scale throughput. This work provides recommendations for Blue Castings to implement mechanisms that reduce rework, improve producibility, and increase throughput to align with Blue Origin’s goals.

A Systematic Political Philosophy of Education

Thu, 01 May 2025 00:00:00 GMT

A Systematic Political Philosophy of Education Pavel, Sonia Maria My dissertation proposes a fundamental repositioning of philosophy of education relative to political philosophy. I argue that we cannot afford doing political philosophy without a theory of education, just as we cannot afford making philosophy of education modular, insulated from the rest of political philosophy. To this end, I propose a systematic political philosophy of education, meaning both a systematization of existing approaches to education and a comprehensive assessment of their merits and limitations. I reconstruct the main theories of education – liberal, conservative, democratic, and critical – from their most basic social ontological assumptions to their political programs for education. I then argue that they all struggle to realize their goals for education either as a result of flawed social ontological assumptions or because of a failure to institutionalize their commitments in practice. The lessons I draw from these critiques form the basis of my own novel systematic theory of education. My theory combines traditional political philosophy with insights spanning critical theory, social ontology, and education studies. The central goal is to reconfigure the school as a democratic institution of social learning that not only enables the flourishing of all students but helps society as a whole progress. The project advances on two levels: a methodological and a substantive-normative one. Methodologically, I resist a growing tendency towards the unmooring of political philosophy and philosophy of education. This tendency is peculiar both from a historical and a conceptual perspective. Historically, education was a core issue of political philosophy. Many, even most, of the canonical political philosophers started from the assumption that education is a central purpose of political life. In my substantive introduction, I take a historical excursus through the canonical political thinkers who best exemplify this emphasis on education: Plato, Rousseau, and Dewey. For all the differences in their views, all three understood education as essential to realizing their visions. They would have regarded any political philosophy that failed to address education as incomplete. Today, however, few political philosophers address the subject at all, let alone give it pride of place in their theories. This unmooring has had bad consequences for both subfields. Much contemporary work in philosophy of education takes for granted a liberal social ontology and liberal normative commitments without sufficient critical scrutiny. Similarly, most contemporary political theory neglects the topic of education and operates under the assumption of fully formed liberal agents. The lack of conceptual clarity is mirrored in political practice. Education is marred by persistent and seemingly intractable disagreements – from controversies about indoctrination to failures to realize the ideal of equality of opportunity. Our substantive disagreements about education, I argue in my first chapter, are not merely value disagreements about the goals of education. They stem from deep-rooted social ontological assumptions about the nature of human beings and society. But these social ontological assumptions are rarely acknowledged, let alone articulated, by political philosophers or philosophers of education. To correct this, I propose a novel metatheory that shows the systematic connections between the social ontology, normative commitments, and political programs of our dominant approaches to education (liberal, conservative, democratic). My reconstruction illuminates several surprising agreements and differences between them. For example, it reveals that many of our most heated political debates about education, between left and right liberals, are merely intramural disagreements among thinkers committed to the same individualist ontology. The systematic reconstruction also illuminates these theories’ failure to generate a coherent vision for education. My critiques show that each approach is characterized by a flawed or incomplete social theory which prevents it from promoting its own values and fulfilling its aims for education. In the case of liberal theories, I show that the liberal goal of cultivating autonomy is selfundermining in light of liberal theory’s individualist social ontology. In the second chapter, I turn to critical theories, which focus on the function of education in reproducing our broader social system. Whereas the dominant approaches start by asking about the nature and goals of education in general, critical theories analyze our contemporary educational systems under specific political and economic conditions. They reveal how schools contribute to perpetuating an oppressive and unjust social system. In other words, the focus of these theories is not on the school as a standalone institution, but as a particularly important subsystem in a larger process of social reproduction. While they are promising in many ways, I nevertheless argue that critical theories of education also have distinct limitations. In particular, even though their social theory and normative commitment are more compelling than the dominant views’, they do not satisfactorily translate these into practical proposals for remaking our systems of education. Having found none of the existing approaches fully satisfactory, I start developing the positive and evaluative dimensions of my own view in the third chapter. I go beyond critical social theory while relying on the broad strokes of its ontology of the human. My aim is to supplement this ontology by drawing on both empirical social studies and complexity theory to more precisely characterize the social relations and practices that constitute the domain of education. More specifically, I argue that we can best understand the educational subsystem by attending to its overlap and co-integration with the family, the state, and economic production. Schools are the mediating institutional domain between the family on one hand and the polity and economic production on the other. At the evaluative level, I articulate three critiques of social pathologies that I believe have been ignored or underutilized in critical education studies: alienation, commodification, and fragmentation. Alienation refers to a pathological relation of disconnection from one’s own learning, other students, and teachers. Commodification and fragmentation, on the other hand, are problems with the organization and distribution of resources in the education system. In my fourth and final chapter, I propose a new program for education that seeks to overcome some of the barriers faced by other systematic theories of education. Attempting to counter the problems I diagnosed and explained in the third chapter, I argue for a few different kinds of interventions. First, I propose restructuring the educational system in order to resist fragmentation by pursuing a unified distributive pool, consolidating school districts, and abolishing charters. Second, I argue for a reconfiguration of the co-integrated subsystems of the family, the school, and production that seeks to empower both children and those involved in their care to be involved in free, meaningful work. Finally, I articulate a set of classroom-level practices that seek to equalize access to development for individual students while cultivating their collective social and political imagination. One of the long-term goals is to make schools into democratic institutions of social learning, through which we strive to remove social blockages such as ideology and reflexivity deficits, in order to collectively solve political problems.

Searching with Intuition: Exploring (the bounds of) LLM-guided Search with Unknown Objectives

Thu, 01 May 2025 00:00:00 GMT

Searching with Intuition: Exploring (the bounds of) LLM-guided Search with Unknown Objectives Kaashoek, Justin H. Large language models (LLMs) can perform a wide range of search and optimization tasks over discrete spaces. This work seeks to explore the limits of LLM-guided search. We construct a set of text optimization tasks with different levels of "intuitiveness'' and evaluate whether LLMs can effectively optimize objectives. We show that the LLM's performance depends not only on its intuition for the objective, but also on the alignment between the objective and its priors. We also find that the LLM can successfully optimize an objective even without an explicit description of the objective. Our results largely focus on greedy search strategies; we develop a theoretical characterization of conditions under which greedy search is optimal, meaning the LLM's failures result from a fundamental inability to take gradient-like steps, not suboptimal search.

A Semantic Account of Distributional Constraints on Temporal in-Adverbials

Fri, 01 Sep 2023 00:00:00 GMT

A Semantic Account of Distributional Constraints on Temporal in-Adverbials Rouillard, Vincent S Temporal in-adverbials (TIAs) are a class of English expressions that can be exemplified with in three days. They are remarkable in that, depending on the syntactic position they occupy, TIAs are subject to very different distributional constraints. In some configurations, their licensing is conditioned by the lexical aspect of verbal predicates. In others, these expressions are negative polarity items. Though both varieties of TIAs have been discussed extensively in the semantics literature (Gajewski, 2005, 2007; Hoeksema, 2006; Iatridou and Zeijlstra, 2017, 2021; Krifka, 1989, 1998), no attempt has been made to understand the relationship between the two. I offer a unified semantic analysis of TIAs, which derives from semantic principles their eclectic distributional constraints.

Metrical Grids and Active Edges

Fri, 01 Sep 2023 00:00:00 GMT

Metrical Grids and Active Edges Asherov, Daniel Theories of word stress assignment differ in the kind of representations they adopt. One family of theories takes stress to be assigned by grouping stress-bearing elements into small units below the level of the word (typically, metrical feet), such that one element in each unit is marked as stronger, hence stressed (e.g., Liberman and Prince 1977; Hayes 1980). Another family of theories, often referred to as grid-only, models stress assignment without appealing to feet or similar bracketed representations above the syllable (Prince 1983; Selkirk 1984; Gordon 2002). While the grid-only approach generates the attested languages with relatively simple representations, it also generates a host of patterns which are very different from those attested in human languages (Hayes 1995; Kager 2012; also see Stanton 2016). This thesis aims to solve a set of overgeneration problems that arise in the grid-only approach. The solution involves three components. The first is a novel class of constraints that are sensitive to word edges but unspecified to the edge they apply to (left or right). The value of this edge, considered the “active” edge, is determined by the ranking between two competing constraints (cf. Richards 2016). The second component involves a specific characterization of alignment constraints and the crucial exclusion of computationally weaker or stronger alternatives. The third component is a set of principled fixed rankings between two classes of constraints. In particular, I propose that constraints sensitive to the active edge systematically outrank constraints that regulate rhythmic alternations (cf. van der Hulst 1997; 2012). The result is a grid-only theory of stress that has a significantly tighter fit to the typology compared to previous theories.

Development of a Throttleable Attitude Control Scheme for Electrospray Propulsion Systems

Thu, 01 Jun 2023 00:00:00 GMT

Development of a Throttleable Attitude Control Scheme for Electrospray Propulsion Systems Harjono, Hanna-Lee Electrospray thrusters have emerged as highly promising propulsion options for small satellites due to their compact size, low weight, and power requirements. These thrusters offer precise, efficient, and scalable attitude control, making them ideal for missions requiring fine adjustments and advanced capabilities such as formation flying and docking maneuvers. However, to fully exploit the potential of electrospray thrusters, control strategies specific to them must be developed. In this work, a parameterized, PID gain-scheduled attitude controller that leverages the unique throttleability of electrospray thrusters is developed and validated. The developed controller is adaptable across operating conditions, as well as electrospray thrust coefficient values. Extensive modeling efforts are undertaken to incorporate the throttleability and operational constraints of electrospray thrusters, ensuring accurate performance predictions. The control system is simulated under various operating conditions to assess and verify its functionality and robustness against disturbance torques. Validation experiments in a magnetic levitation CubeSat testbed are proposed.

Visibility in synthetic aperture radar satellite data: metric formulation, observation scheduling, and orbit design

Thu, 01 May 2025 00:00:00 GMT

Visibility in synthetic aperture radar satellite data: metric formulation, observation scheduling, and orbit design Kramer, Evan L. Earth observation satellites serve as vital information gatherers for effectively addressing some of humanity’s most pressing challenges including management of limited resources and minimization of losses from disasters. Synthetic aperture radar (SAR) is a type of active remote sensing instrument that operates in the microwave portion of the electromagnetic spectrum and is a preferred Earth observation system thanks to the reliable imagery it can collect in all illumination and weather conditions. SAR data is acquired using a side-looking viewing geometry in which the radar is pointed perpendicular to the satellite platform’s direction of motion. This viewing geometry, in conjunction with the illuminated terrain’s topography, results in geometric distortions termed layover and shadow. These distortions degrade the utility of the collected imagery since they effectively obscure portions of the image and preclude extraction of actionable insights. While geometric distortions will be everpresent in SAR imagery, their location and coverage can be manipulated by controlling the relative orientation between the satellite and the illuminated topography. Such manipulation has historically been infeasible for legacy SAR satellites that collect globally consistent data sets under rigid operating requirements. However, the recent advent of commercial SAR satellite constellations has re-framed the practicality of carefully tuned observation geometries that maximize region of interest visibility. Commercial SAR constellations operate on a task-wise basis that grants data end-users flexibility in specifying desired observation parameters including acquisition times and observation geometries. However, a mismatch between on-orbit capabilities and delivered data quality exists due to a lack of formalized tools for planning observations with maximum region of interest visibility. Specifically, no systematic method for identifying visibility-favorable observation geometries exists. This dissertation addresses this gap in a stepwise approach. First, an extension of opensource radar processing software is developed that enables prediction of layover and shadow in a 2D distortion mask for any satellite-target relative geometry. Visibility metrics are then defined to represent the favorability of a particular observation geometry with respect to a distortion mask. The computation of visibility metric scores at geometries spanning the entire sample space enables creation of visibility maps that completely characterize the visibility characteristics of a given region of interest. To broaden the suitability of visibility maps for observation planning, a set of generalizable visibility maps are created to enable estimation of region of interest visibility characteristics in mission scenarios that are computationallyconstrained and information-limited. Visibility maps are then directly integrated into satellite operations by developing the first SAR observation scheduling algorithm that explicitly accounts for visibility. Finally, visibility is considered in the orbit design process to establish general guidance on optimal repeat ground track orbit parameters for pre-defined region of interest visibility characteristics. Region of interest visibility improvements of up to 90% are obtained for individual tasks when using the observation planning tools developed in this dissertation. Constellation-wide visibility improvements of 18% are achieved with modest reductions in traditional performance measures when integrating visibility into observation scheduling. Two-fold improvements in the visibility characteristics of observation opportunities are attained for orbits designed to maximize overpass geometry quality. The contributions of this dissertation are timely, given the concurrent proliferation of flexible, high-resolution SAR observation capabilities, and lay the groundwork for enabling the acquisition of SAR data that is maximally useful for limited resource management, disaster response, and other applications.

Simulation Modeling of Drug Substance Tech Transfer Timelines at Amgen

Thu, 01 May 2025 00:00:00 GMT

Simulation Modeling of Drug Substance Tech Transfer Timelines at Amgen Goel, Viraat Yogi Technology transfer (TT), or the process by which a product's manufacturing is moved and scaled, is a complex business process with countless deliverables and stakeholders. This is especially true in biomanufacturing, where drug commercialization timelines are measured in years, manufacturing facilities are specially designed, and regulations must be stringently met. This systems-level complexity can create inefficiencies in the TT process, lengthening timelines and wasting resources. In this project, we use simulation modeling techniques to digitally model Amgen's Commercial Tech Transfer (CTT) process for biologic drugs. We use virtual experimentation to identify key bottlenecks in the TT workflow, quantify how workstream alterations impact project timelines, and identify process changes likely to shorten timelines. We also extend this analysis to Amgen's New Product Introduction (NPI) process, identifying how coordination between upstream and downstream processes may accelerate NPI timelines. Finally, we link this project to the ongoing development of TT data visualization dashboards.

Safety Stock Modeling for a Medical Devices Supply Chain

Thu, 01 May 2025 00:00:00 GMT

Safety Stock Modeling for a Medical Devices Supply Chain Chong, Julie This thesis examines the current inventory management practices at a leading manufacturer of medical devices, and identifies areas for significant improvement. The analysis reveals inefficiencies in safety stock management, with finished goods inventories being excessively high and raw material stocks being underestimated. The study applies single-echelon and multi-echelon inventory modeling to demonstrate potential cost savings through optimized safety stock levels. Additionally, it highlights the importance of reevaluating high service level targets and improving forecasting accuracy to reduce reliance on costly countermeasures. The thesis also emphasizes the need for effective management of component lead times and enhanced data visibility. Recommendations include transitioning to data-driven safety stock calculations, adopting multi-echelon inventory optimization, reassessing service level targets, enhancing forecasting accuracy, and improving component lead time management. By implementing these strategies, the company can enhance operational efficiency, reduce costs, and build greater resilience in its supply chain.

Expanding home broadband coverage through existing Low Earth Orbit megaconstellations

Thu, 01 May 2025 00:00:00 GMT

Expanding home broadband coverage through existing Low Earth Orbit megaconstellations Gonzalez Martinez, Gretel Expanding broadband access to underserved areas continues to be a significant challenge for Internet Service Providers (ISPs). While its services perform well in high-density regions, they face scalability limitations in sparsely populated areas where infrastructure costs must be spread across a smaller customer base. This study explores the potential of Low Earth Orbit (LEO) satellite megaconstellations as a scalable solution for extending broadband coverage in the United States. By analyzing the technical capabilities, deployment timelines, and economic feasibility of partnering with LEO satellite providers, this research offers a strategic framework for integrating satellite broadband into ISPs service portfolio. A customer demand model identifies approximately 17 million unserved households within the addressable market of one of the largest U.S. telecommunications companies. The business case assessment evaluates broadband profitability by optimizing customer base size relative to proximity to existing infrastructure. While fiber optics remains the most profitable solution in high-density areas and fixed wireless access effectively utilizes excess 5G capacity, both require substantial infrastructure investment, limiting their feasibility for rural broadband expansion. In contrast, a satellite broadband partnership emerges as the most cost-effective solution for at least 1 million households, surpassing the profitability of currently existing offerings. With minimal capital investment, satellite technology enables rapid customer acquisition and scalable nationwide expansion. The analysis highlights that wholesale agreements play a critical role in profitability and the need to secure a minimum revenue share of 16.5% to reach the break-even point. Performance modeling and curve approximation techniques estimate that if Kuiper meets Federal Communications Commission (FCC) deployment milestones, it could serve 8.5 million customers by 2026, with full nationwide coverage projected by 2029. Under a 200x oversubscription model, Kuiper’s total subscriber capacity could scale to 32.8 million, demonstrating its ability to complement current broadband o!erings. While LEO broadband networks can achieve capacities in the tens of Tbps, they remain far below fiber networks, which operate in the thousands of Tbps. Rather than competing directly, satellite broadband is positioned as a complementary solution, addressing connectivity gaps in rural and underserved regions. To capitalize on these findings, this study recommends leveraging existing LEO megaconstellations to expand broadband coverage nationwide. A phased rollout should begin with a beta program in California, the state with the highest number of unserved households, to validate network performance and optimize deployment for broader expansion. Partnering with an existing LEO megaconstellation could e!ectively bridge the digital divide in rural areas, expand service offerings, and enable a stronger position in the growing satellite broadband market.

Economic Determinants of Increased Use of Performance-Vesting Provisions in CEO Incentives

Thu, 01 May 2025 00:00:00 GMT

Economic Determinants of Increased Use of Performance-Vesting Provisions in CEO Incentives Kim, Jason Gwanhee This study examines the determinants of firms adopting performance-vesting long-term incentive (PLI) awards, a rapidly growing form of executive compensation. Using data provided by Equilar on Russell 3000 firms, I investigate how a firm's contracting environment and inter-firm networks influence the adoption and design of PLI awards. I find that stock liquidity and analyst coverage significantly increase the likelihood of adoption by enhancing the informativeness of performance measures. The findings suggest that firms adopt PLI awards to better align managerial incentives with shareholder interests, focusing on the measures that are both reliable and strategically aligned. I also show that board interlocks, particularly those involving compensation committee members, and shared compensation consultants play a significant role in facilitating the diffusion of PLI awards across firms.

Essays on Content Moderation Interventions for Addressing Online Misinformation

Thu, 01 May 2025 00:00:00 GMT

Essays on Content Moderation Interventions for Addressing Online Misinformation Martel, Cameron In Chapter 1, I examine the efficacy of fact-checker warning labels as a content moderation intervention for addressing online misinformation. Warning labels from professional fact-checkers are one of the most historically used interventions against online misinformation. But are fact-checker warning labels effective for those who distrust fact-checkers? In a first correlational study, we validate a measure of trust in fact-checkers. Next, we conduct meta-analyses across 21 experiments in which participants evaluated true and false news posts and were randomized to either see no warning labels or to see warning labels on a high proportion of the false posts. Warning labels were on average effective at reducing belief in, and sharing of, false headlines. While warning effects were smaller for participants with less trust in fact-checkers, warning labels nonetheless significantly reduced belief in, and sharing of, false news even for those most distrusting of fact-checkers. Our results suggest fact-checker warning labels are a broadly effective tool for combatting misinformation. In Chapter 2, joint with Jennifer Allen, Gordon Pennycook, and David G. Rand, I investigate the potential of crowdsourced fact-checking systems to identify misleading online content. Social media platforms are increasingly adopting crowd-based content moderation interventions for identifying false or misleading content. However, existing theories posit that lay individuals can be highly politically biased, and that these strong political motivations inherently undermine accuracy. Alternatively, we propose that political and accuracy motivations may, in some cases, operate in tandem – in which case politically motivated individuals need not hamper truth discernment. We empirically assess this by analyzing a survey study of misinformation flagging and field data from X’s Community Notes. Consistent with a simple model of flagging behavior, posts that are both false and politically discordant are flagged the most. Importantly, we find that more politically motivated users flag a greater number of posts, engage in more politically biased flagging, and yet exhibit the same or better flagging discernment. Together, these results show that politically motivated individuals are integral to provisioning a high overall quantity and quality of crowdsourced fact-checks. In Chapter 3, I assess the perceived legitimacy of different content moderation interventions for addressing online misinformation. Current content moderation practices have been criticized as unjust. This raises an important question – who do Americans want deciding whether online content is harmfully misleading? We conducted a nationally representative survey experiment in which U.S. participants evaluated the legitimacy of hypothetical content moderation juries tasked with evaluating whether online content was harmfully misleading. These moderation juries varied on whether they were described as consisting of experts, laypeople, or non-juries. We also randomized features of jury composition (size, necessary qualifications) and whether juries engaged in discussion during content evaluation. Overall, participants evaluated expert juries as more legitimate than layperson juries or a computer algorithm. However, modifying layperson jury features helped increase legitimacy perceptions – nationally representative or politically balanced composition enhanced legitimacy, as did increased size, individual juror knowledge qualifications, and enabling juror discussion. Our findings shed light on the foundations of institutional legitimacy in content moderation and have implications for the design of online moderation systems.

Materials and Devices for Optoelectronic Packaging

Sat, 01 Feb 2025 00:00:00 GMT

Materials and Devices for Optoelectronic Packaging Weninger, Drew Michael Over the last two decades, improvements in semiconductor manufacturing have allowed for the commercialization of silicon photonic integrated circuits with over 10,000 devices. These chips are critical to data and telecommunications networks where they convert and encode optical signals to electrical signals, and vice versa, in the form of transceivers. Scaling up the number of transceivers and optical fiber connections, or optical input/output (I/O), will be critical to meet the exponential rise in demand for cloud data capacity since 2010. However, the costly process of active alignment and bonding of optical fiber arrays directly to photonic chips presents a barrier to their high volume packaging and assembly. This approach limits optical I/O density to a maximum of 8 connections per millimeter since optical fibers for communications applications have cladding diameters of 125 micron. To address this challenge, this thesis explored a new field of silicon integrated photonics involving chip-to-chip (i.e. flip-chip) optical coupling. Evanescent chip-to-chip couplers were designed, fabricated, packaged, and tested for directly connecting silicon photonic chips to other silicon photonic chips, interposers, or printed circuit boards using automated assembly. The design's compact footprint allows for coupler pitches below 10 micron, or an optical I/O density of greater than 100 connections per millimeter, to be realized - an order of magnitude improvement over fiber-to-chip connections. By designing the coupler to use silicon materials and back-end-of-line compatible packaging processes, ease of integration with existing microelectronic foundry tool sets was ensured. Results from an experimental flip-chip coupler prototype showed greater than 90% coupling efficiency with micron scale alignment tolerances when coupling from silicon nitride to silicon-on-insulator waveguides, the first demonstration of such a device. To further improve optical flip-chip coupler performance, designs were proposed for combining the evanescent coupler with an integrated graded index lens using silicon oxynitride films. Such a device would provide a universal coupling interface in silicon photonics for both chip-to-chip or fiber-to-chip connections. Simulations showed sub-dB coupling loss across all interfaces including flip-chip coupling across a 10 micron gap. Initial fabrication processes were established to deposit, pattern, and etch greater than 10 micron thick silicon oxynitride graded index lenses on silicon and glass substrates. In showing that automated pick-and-place tools can be used for photonic chip assembly, this work represents a critical step in eliminating active alignment and sustainably scaling optical I/O in future transceiver packages.

Selecting for and Selecting Despite: A Javanese case study

Fri, 01 Sep 2023 00:00:00 GMT

Selecting for and Selecting Despite: A Javanese case study Lesure, Cora This is an investigation of the argument structure of Javanese (Austronesian, Indonesia) which focuses on the distribution of four core derivational morphemes: the Actor Voice prefix, and the suffixes -ake, -i, and -an. The project is based on original consultant work conducted with a speaker of the Central dialect of Javanese. The work establishes language internal diagnostics for various aspects of a stem's lexical semantics and lexical category and then utilizes these criteria to analyze a wide variety of morphological derivatives, both verbal and nominal. The resulting analysis is able to predict the distribution of derivational morphemes and the nature of their resulting derivatives to a higher degree than what was previously understood to be possible.

Uncovering Mandarin Speaker Knowledge with Language Game Experiments

Fri, 01 Sep 2023 00:00:00 GMT

Uncovering Mandarin Speaker Knowledge with Language Game Experiments Fu, Boer Mandarin Chinese offers many intriguing puzzles for linguists because it has a shortage of morphophonological alternations. This has resulted in indeterminacy in various aspects of its phonological grammar, triggering much debate on syllable structure and allophonic mapping. The ambiguity of the data is also a problem for children acquiring Mandarin since alternative grammars can account for the surface forms equally well. In order to find out what Mandarin speakers have learned about the phonology of their language, I conducted two language game experiments based on fanqie secret languages. It was found that markedness and faithfulness constraints are psychologically real for Mandarin speakers. Furthermore, the interactions between markedness and faithfulness constraints are shown to have an effect on glide movement in the language game. In addition, much speaker variation was observed in the experiment. I demonstrate that it is the result of constraint ranking variation. Nevertheless, general population-level trends on constraint ranking could still be identified. These trends lead to insights on phonological learning beyond Mandarin, showing evidence for naturalness bias and lexicon optimization.

Business Value of Enterprise Digital Architecture

Sun, 01 May 2022 00:00:00 GMT

Business Value of Enterprise Digital Architecture Venkata Aditya, Saraswatula (Adi SV) Digital technologies are fundamentally reshaping markets and organizations globally. This thesis is exploratory research that seeks to explain how large multi-regional and global enterprises determine, prioritize, measure, and manage business value outcomes of digital investments over time. I examine the value construct of digital initiatives in firms from different industries by interviewing various stakeholders. Insights surfaced from this primary research are analyzed in conjunction with the concepts from current literature. Qualitative findings are proposed, and a list of value metrics is presented that can serve as a future reference for firms. A causal loop diagram is proposed to visualize firm capabilities and value dynamics.

Risk-Aware Reinforcement Learning with Safety Constraints

Thu, 01 May 2025 00:00:00 GMT

Risk-Aware Reinforcement Learning with Safety Constraints Feng, Meng Safety is a critical concern in reinforcement learning (RL) and learning-based systems more broadly, as ensuring reliable and safe decision-making is essential for their deployment in real-world applications. Traditional approaches to address safety often rely on techniques such as reward shaping, carefully curated training data, or explicit handcrafted rules to avoid unsafe actions. More recent advancements have adopted the Constrained Markov Decision Process (CMDP) framework, which trains agents while explicitly enforcing constraints on auxiliary measures such as safety or risk. However, these methods often suffer from significant constraint violations. This thesis identifies the root cause of such violations as stemming from the pursuit of maximal task performance in each policy update. Given the inherent limitations of sample-based constraint assessments in RL, where data is limited and approximation errors are inevitable, these methods often fail near constraint boundaries, leading to excessive violations. To address this, we propose a novel constrained reinforcement learning algorithm that dynamically adjusts its conservativeness during policy updates. By incorporating the risk of constraint violation into the update process, our method can shift focus toward constraint satisfaction when violations are likely, while still striving to improve task performance whenever feasible. Our algorithm reduces constraint violations by up to 99% compared to state-of-the-art baselines while achieving comparable task performance. In the second part of this thesis, we extend CMDPs to address multi-goal, long-horizon problems. We augment the CMDP formulation to incorporate goals, enabling it to handle multiple goals while preserving goal-independent constraint specifications in CMDP. To tackle the complexity of long-horizon tasks with high-dimensional inputs (e.g., visual observations), we propose a method that integrates planning with safe reinforcement learning. By leveraging deep reinforcement learning, we acquire the essential components for planning, including a low-dimensional state-space representation and planning heuristics. The planning algorithm then decomposes long-horizon problems into a sequence of shorter, easier subgoal-reaching tasks. The learned agents safely navigate toward these subgoals step by step, ultimately reaching the final goal. We evaluate our method on both single-agent and multi-agent tasks. In 2D navigation, our approach demonstrated up to 74.2% risk reduction, while in visual navigation, it achieved up to 49.3% risk reduction, all while reaching comparable or better success rates.

Optimizing Inventory Rebalancing: Strategies for Managing Excess Inventory in a Dynamic Supply Chain

Thu, 01 May 2025 00:00:00 GMT

Optimizing Inventory Rebalancing: Strategies for Managing Excess Inventory in a Dynamic Supply Chain Oludipe, Lanre The increasing demand for faster consumer delivery has led retailers to establish smaller regional distribution centers alongside traditional main distribution centers (MDCs). However, the limited capacity of some of these regional centers heightens the need for precise inventory forecasting and deployment to minimize excess inventory, particularly when few viable outlets exist for excess inventory. This research examines strategies to mitigate excess inventory at regional centers through inventory rebalancing, the integration of additional outlets, and modifications to existing inventory policies. A Monte Carlo simulation was conducted to compare the performance of the current system with a modified system incorporating these enhancements. The results showed that the modified system improved capacity utilization and reduced inventory deployment from the MDC without affecting margin. These improvements can enable more agile operations at smaller regional centers, reduce inventory buildup, and reduce the pressure of precise inventory deployment.

Electric Vehicle Fleet Charging: A Simulation-Based Comparison of Charging Strategies and Cost Implications

Thu, 01 May 2025 00:00:00 GMT

Electric Vehicle Fleet Charging: A Simulation-Based Comparison of Charging Strategies and Cost Implications Knapp, Rachael The global shift to electric vehicles (EVs) is progressing rapidly, driven by the need to reduce greenhouse gas (GHG) emissions and global reliance on fossil fuels. However, fleet electrification presents unique challenges, particularly in regard to rolling out the necessary charging infrastructure and operational efficiency. This study examines how various depot-based fleet charging strategies impact up-front capital and long-term operational expenditures. The operational feasibility of each method is evaluated through the use of a discrete event simulation. The study incorporates fleet data to assess the time required to charge the fleet, the number of chargers needed, and the number of associates needed to operate manual strategies. The analyzed charging methods include dedicated level 2 charging, vehicle swapping, level 2 cable swapping, level 3 cable swapping, sequential and simultaneous charging. Key findings indicate that while a 1:1 vehicle-to-charger ratio ensures charging reliability within the designated time, it incurs the highest capital costs. Alternative strategies, such as cable swapping and simultaneous charging, significantly reduce costs while successfully charging the fleet within the charging window.

An Integrated Optimization Model for Large-Scale EV Fleet Deployment: Balancing Emissions Reduction and Operational Costs

Thu, 01 May 2025 00:00:00 GMT

An Integrated Optimization Model for Large-Scale EV Fleet Deployment: Balancing Emissions Reduction and Operational Costs Kasliwal, Mohit This thesis presents an integrated optimization framework designed for the large-scale deployment of electric vehicles (EVs) within commercial fleets, specifically focusing on balancing emissions reduction and operational cost efficiencies. Utilizing Verizon’s extensive fleet of over 10,000 light-duty vehicles across 1,000 sites as a case study, the research addresses the challenges and complexities in effective site selections for such a large and dispersed fleet. The research involved developing and testing several optimization models under varying scenarios, including scenarios prioritizing maximum operational savings, maximum emissions reduction, and a hybrid model employing an internal cost of carbon (ICC) to balance both operational and environmental objectives. The model essentially develops a ranking system for sites – suggesting which sites to electrify in which year and order, with how many EV conversions (from existing ICE vehicles) at each site. The results highlight the importance of tailoring EV deployment strategies to site-specific conditions, such as unique vehicle usage patterns, grid emissions profiles, regional operational costs, and available incentives. Particularly, smaller sites were found to offer greater relative benefits in terms of both cost savings and emissions reductions per unit of capital invested due to their high average mileage, making them strategic priorities for early electrification. Operational feasibility was also thoroughly examined, recommending practical constraints such as limiting the number of sites electrified annually to ensure project manageability and effectiveness. Sensitivity analyses addressed critical uncertainties such as battery degradation over the vehicle lifespan and the impact of extreme weather on EV performance. These analyses underscore the necessity of conservative battery range buffers ("safe ranges"). Robust load management strategies can be deployed to significantly reduce demand charges and optimize charging schedules based on time-of-use rates where available. Recommendations from the study advocate for implementing a hybrid optimization strategy incorporating an ICC based on corporate goals, continuous adaptive management informed by ongoing data collection, and strategic infrastructure investments to future-proof EV deployments. Policy alignment is also critical to enhance economic viability via incentives and ensure regulatory compliance. Finally, the thesis proposes future research directions, including investigation of advanced load management and integration with renewable energy sources, exploring bi-directional charging to add revenue streams, incorporating marginal operating emissions rate (MOER) data to further reduce grid emissions and exploring the resilience of EV fleets to power outages. These initiatives aim to further enhance strategic decision-making and ensure the long-term sustainability and efficiency of fleet electrification programs.

Breaking the Chain: Building Resilience in the Insurance Value Chain

Thu, 01 May 2025 00:00:00 GMT

Breaking the Chain: Building Resilience in the Insurance Value Chain Chuah, Chung Jin This thesis examines how strategic transformation approaches reshape the resilience of the Property & Casualty (P&C) insurance industry in the light of ongoing technological disruption, climate change, and regulatory pressures. Through empirical analysis of 9 insurers, the study reveals that while all transformation types improve performance, phased 'test-refine-execute' strategies achieve superior outcomes by combining operational focus with strategic agility. The research identifies four implementation levers: (i) digital modernization, (ii) phased transformation execution, (iii) resource-allocation agility, and (iv) aligned leadership - which together explain why some transformations succeed where others fail."

Domain Adaptation of VLM for Soccer Video Understanding

Thu, 01 May 2025 00:00:00 GMT

Domain Adaptation of VLM for Soccer Video Understanding Jiang, Tiancheng(Tony) Vision Language Models (VLMs) have demonstrated strong performance in multi-modal tasks by effectively aligning visual and textual representations. However, most video under- standing VLM research has been domain-agnostic, leaving the understanding of their transfer learning capability to specialized domains underexplored. In this work, we address this by exploring the adaptability of open-source VLMs to specific domains, and focusing on soccer as an initial case study. Our approach uses large-scale soccer datasets and LLM to create instruction-following data, and use them to iteratively fine-tune the general-domain VLM in a curriculum learning fashion (first teaching the model key soccer concepts to then question answering tasks). The final adapted model, trained using a curated dataset of 20k video clips, exhibits significant improvement in soccer-specific tasks compared to the base model, with a 37.5% relative improvement for the visual question-answering task and an accuracy improvement from 11.8% to 63.5% for the downstream soccer action classification task.

Minimizing Cost of Intra-Yard Finished Vehicle Logistics Through Automation and Optimization

Thu, 01 May 2025 00:00:00 GMT

Minimizing Cost of Intra-Yard Finished Vehicle Logistics Through Automation and Optimization Garber, Jeremy This thesis analyzes and validates autonomous Finished Vehicle Logistics (FVLa) operations, at the plant of an automative Original Equipment Manufacturer (OEM), through the development of a Vehicle-Plug-In (VPI) system with Level 4 autonomous driving capabilities. The research combines process flow analysis with FlexSim simulation modeling to optimize operational parameters and assess safety performance. Results demonstrate FVLa operational feasibility with a recommended VPI inventory of 750 units and 6-hour replenishment cycle. The study's key contributions include a validated operational model using Economic Order Quantity calculations and a safety framework utilizing Bayesian Networks, establishing foundations for the planned 2028 implementation while maintaining required throughput rates and safety standards.

Decarbonized Cement Manufacturing via Advanced Production Technologies

Thu, 01 May 2025 00:00:00 GMT

Decarbonized Cement Manufacturing via Advanced Production Technologies Norwalk, Michael Cement production is the second-largest source of industrial carbon dioxide emissions world-wide. Due to the chemical reactions inherent in its production and the temperatures required to drive those reactions, cement is considered a “hard-to-decarbonize” industry. In this study, three emerging technologies to reduce the carbon intensity of industrial processes, namely, direct high-temperature electric process heat, electric process heat utilizing thermal storage, and liquid amine-based carbon capture are assessed in the context of a greenfield cement production facility relative to a new-build conventional cement plant fueled with natural gas. Cement plants utilizing this set of technologies were modeled in five U.S. geographies to determine the relative economic returns. The economics were assessed, inclusive of available economic incentives, both for the scenario in which the cement produced is sold in the U.S. market and for the scenario in which the cement produced is exported to the European Union (E.U.) market to assess potential benefits from the E.U. carbon pricing system. The analysis indicates that at current technology prices, the economic returns of the assessed technologies, while in some cases profitable, continue to lag those of conventional production technology for the domestic U.S. market. As costs come down as technology is deployed, the economics of carbon capture solutions have the potential to be competitive with conventional technology. The E.U. carbon emissions penalties are effective in altering the economics in such a way that implementing carbon capture systems becomes the most attractive economic option, demonstrating the power of carbon emissions markets. With increased technology deployment as well as the adoption of targeted incentives in the U.S. market, the adoption of low carbon cement production technologies can be accelerated.

Obscured universality in Mandarin

Fri, 01 Sep 2023 00:00:00 GMT

Obscured universality in Mandarin Chen, Fulang In this dissertation, I investigate the apparently distinctive syntactic properties associated with the BEI-construction, the BA-construction, and resultative constructions in Mandarin Chinese, which I argue obscure properties that are universal across natural languages. In the case of the Mandarin BEI-construction, it exhibits both passive-like and tough-movementlike properties. I argue for a novel analysis of the BEI-construction as a passive construction, where the passive head/BEI hosts a composite probe [𝜙 + Ā], which triggers composite A/Ā-movement, in the sense of Van Urk (2015). The subject in the BEI-construction is derived via (successivecyclic) composite A/Ā-movement, followed by a terminating step of A-movement, similar to Longenbaugh’s (2017) analysis of English tough-movement. Under the proposed analysis, the mixed A/Ā-properties associated with the BEI-construction are direct consequences of composite A/Āmovement (following Van Urk 2015; Longenbaugh 2017). In the case of the Mandarin BA-construction, it involves an apparently pre-posed noun phrase (the post-BA NP) with an affectedness interpretation, which can be identified with either the subject of a resultative phrase in a complex predicate or the direct object of a simple transitive verb. I argue for a novel analysis of the Mandarin BA-construction as a causative construction, where the causative head, which selects a predicate of the caused/resulting event and projects a predicate of the causing event (following Pylkkänen 2002, 2008), has two additional arguments: a causer and a causee. The post-BA NP, as the causee argument of the causative head, also controls a PRO subject in a resultative phrase (following Huang 1992), which is overt in a complex-predicate BAconstruction and is phonologically null in a simple-transitive BA-construction (following Sybesma 1992, 1999). The post-BA NP is interpreted as being affected in the causing event, in the sense that it is caused to perform an action or undergo a change of state (following Alsina 1992). Lastly, in Mandarin, there is no apparent unaccusative-unergative distinction in resultative constructions, unlike languages like English, where distinctions between resultative constructions with unaccusative and unergative matrix verbs follow from the Unaccusativity Hypothesis (Perlmutter 1978; Burzio 1986) and general principles such as the Direct Object Restriction (Simpson 1983; Levin & Rappaport Hovav 1995) and Burzio’s generalization (Burzio 1986). I argue that resultative constructions in Mandarin are causative constructions, where the causative head has four possible argument structures, depending on whether the matrix verb is unaccusative, unergative, or transitive, as well as the semantic relation between the matrix subject and the matrix verb (and between the post-verbal NP and the matrix verb). Despite the fact that the argument structure of the causative head obscures the argument structure of the matrix verb, I argue that in Mandarin resultative constructions, the sole argument of an unaccusative matrix verb is always a causee argument, whether or not an additional causer external argument is present, while the sole argument of an unergative matrix verb, which is a causer external argument otherwise, is a causer argument when the causer is an internal argument. The dissertation showcases how Mandarin provides insight in defending and expanding our knowledge of cross-linguistic properties such as passivization (which embodies Burzio’s generalization and feature-driven movement), composite probing, the bi-clausal syntax and bi-eventive semantics of causative constructions, as well as the nature of affectedness (in causative constructions) and implications for the Unaccusativity Hypothesis and the Uniformity of Theta-Assignment Hypothesis (Baker 1988).

Single-Polarity Ion Electrospray Propulsion

Thu, 01 Jun 2023 00:00:00 GMT

Single-Polarity Ion Electrospray Propulsion Shaik, Saba Zareen Electrospray thrusters are highly efficient spacecraft propulsion devices that accelerate ions sourced from ionic liquid propellants to produce thrust. Typically, electrosprays are fired in a dual-polarity configuration in which the polarity of the ion beam is periodically reversed. This strategy is difficult to implement and imposes limitations on system size and performance. We instead propose a single-polarity design where negative ions are emitted continuously from the thruster, enabling extreme miniaturization, faster startup, better emission stability, and simpler power processing. This thesis investigates two challenges associated with the single-polarity design. First, system lifetime is of principal importance for electrospray propulsion systems in general and must be verified for a single-polarity implementation. Long-duration electrospray tests are performed, demonstrating that single polarity thrusters achieve comparable lifetimes and performance to state of the art systems with high mass utilization and minimal hardware degradation. An additional challenge is propellant electrochemistry, triggered when positive counterions accumulate in the ionic liquid. A suite of experiments is conducted to identify and characterize electrochemical processes, including electrical double-layer potential evolution and gas-phase product formation, in electrospray thrusters over long firing durations.

Comparative Analysis of Semiconductor Investment Environments in the U.S. and China

Thu, 01 May 2025 00:00:00 GMT

Comparative Analysis of Semiconductor Investment Environments in the U.S. and China Zhang, Hanxue Semiconductors are fundamental to Artificial Intelligence(AI) and central to global technological competition. Against this backdrop, this thesis compares semiconductor primary investment environments in the United States and China, examining their implications for industry development and innovation. The study employs a mixed-methods approach, combining expert interviews, data analysis, and natural language processing (NLP). It draws on primary market investment, M&A deals and government grants data to examine capital structures, investment stages, sectoral focus, and exit efficiency. Furthermore, it analyzes nearly 3,000 semiconductor industry reports(2020-2025) to identify evolving strategic priorities and thematic trends shaping these environments. Findings reveal that China’s state-led, vertically integrated model prioritizes upstream capacity building and supply chain autonomy, supported by government guidance funds, private capital, and policy-driven mechanisms. However, there remains a significant gap in leading-edge chips, necessitating precise investments and patient capital to bridge this divide. While the U.S. ecosystem, shaped by major technology firms and federal support, focuses on design innovation and cutting-edge technologies. However, structural constraints such as limited exit pathways, fragmented fabrication capacity, and insufficient industrial policies may hinder the U.S. in nurturing innovation-driven small and medium-sized enterprises (SMEs) in the semiconductor industry. This thesis highlights the structural divergence between the U.S. and China’s semiconductor ecosystems by examining policy, primary market capital, and investment dynamics. It offers policymakers and investors a strategic overview of how these forces shape innovation and resilience, while identifying emerging investment priorities and future development paths.

Forecasting Automotive Production Volume Using Regression and Time Series Modelling

Thu, 01 May 2025 00:00:00 GMT

Forecasting Automotive Production Volume Using Regression and Time Series Modelling Gong, Yutao Accurate forecasting of automotive production volumes is a critical capability for suppliers navigating an increasingly volatile industry. Overly optimistic forecasts, particularly from Original Equipment Manufacturers (OEMs), lead to misallocated capacity and lost opportunities across the supply chain. This thesis investigates whether advanced statistical models can improve upon benchmark industry forecasts and provide automotive suppliers with more reliable, practical tools for demand planning. Several forecasting methodologies are evaluated, including ARIMA, standard linear regression, Lasso regression, Theta model, and a hybrid Boosted Theta model. Models are tested across North America, Europe, and Greater China using 2000-2024 vehicle production and macroeconomic data. Results show that Theta model outperforms industry forecasts across both 1-year and 5-year horizons in North America and Europe. Its simplicity, low data requirements, and robustness to market volatility make it suitable for industrial use. The model was successfully implemented at Commonwealth Rolled Products, an aluminum rolling mill in Kentucky, portfolio company of American Industrial Partners (AIP), where it was adopted for 2025 planning and drove a shift towards data-centric forecasting practices. This research presents a real-world example of applying academic techniques to solving actual business problems, serving as a valuable reference for suppliers seeking to improve forecast accuracy and operational planning in the evolving automotive landscape.

The role of university venture funds in supporting early-stage Japanese startups

Thu, 01 May 2025 00:00:00 GMT

The role of university venture funds in supporting early-stage Japanese startups Brillaud, Nami This thesis explores how university venture funds in Japan are uniquely positioned to turn the country’s innovation capacity into entrepreneurial capacity by supporting early-stage startups. While Japan consistently ranks high in research output, much of this potential is not being translated into successful entrepreneurship. Risk capital is scarce compared to other ecosystems, particularly for deep tech, and support systems for early-stage startups are still limited. University venture funds – which inherently connect universities, entrepreneurs, and risk capital – are well positioned to bridge this gap. Yet despite their growing relevance, their evolving role in supporting Japanese early-stage startups is understudied. This study compares university venture funds with different profiles – ranging from leading and longstanding funds like UTEC, to public-private venture funds established through government initiatives, to recent funds with diversified structures – analyzing how they are structured, how they invest, and what results they have seen so far. It then builds on startup examples and interviews with university venture funds to identify how these funds can better support early-stage startups through improved fund operations, stronger pre-seed support, as well as a strategic approach to growth and exits. Ultimately, this thesis advocates for actionable solutions informed by global practices but adapted to Japan’s unique startup ecosystem.

Analyzing Procurement Data for Cost Saving Application

Thu, 01 May 2025 00:00:00 GMT

Analyzing Procurement Data for Cost Saving Application Pan, Haoting In an increasingly data-driven business environment, procurement analytics plays a critical role in optimizing costs and improving supply chain efficiency. This thesis examines the development and implementation of the Lifecycle Cost Management (LCM) tool at Caterpillar Inc., a global leader in heavy equipment manufacturing. Given Caterpillar's decentralized procurement structure, managing cost-saving initiatives across its 150 facilities (Caterpillar | Caterpillar Frequently Asked Questions (FAQs), n.d.) and 28,000 suppliers (Caterpillar | Caterpillar at a Glance, n.d.) poses a significant challenge. The LCM tool leverages machine learning models to identify overpriced purchase orders (POs) and generate actionable cost-saving opportunities. This study explores the methodology used to enhance LCM's predictive capabilities, including data sourcing and cleaning, feature engineering, model selection, and validation. Various regression models, clustering techniques, and machine learning algorithms, such as Random Forest and XGBoost, are tested to identify cost outliers. A validation process is implemented to ensure that flagged outliers are cost-saving opportunities appropriate for execution. Beyond technical development, the thesis addresses the processes of digital tool adoption within Caterpillar’s procurement teams. A change management approach is employed, incorporating buyer interviews, stakeholder engagement, and iterative user experience (UX) improvements. Through case studies, the study highlights the machine learning model performance and tangible financial impact of LCM. The LCM tool has identified more than $100M data-driven potential savings, and hopes to realize 20% of the savings. Because Caterpillar’s procurement contracts are often long-term, these savings can be considered perpetual. Findings indicate that while machine learning models effectively identify cost outliers, their success is contingent on robust data governance, stakeholder buy-in, and integration into procurement workflows. The study underscores the importance of data management, organizational alignment, and continuous refinement of digital procurement tools. Future work recommendations are enhancing data infrastructure, integrating AI-driven contract management and analysis, and refining cost estimation methodologies. The insights gained contribute to the broader application of procurement analytics and digital transformation in manufacturing enterprises.

Impact Evaluation and Prioritization Framework for Manufacturing Inspection Technology Investment

Thu, 01 May 2025 00:00:00 GMT

Impact Evaluation and Prioritization Framework for Manufacturing Inspection Technology Investment DiDio, Isabella Advancements in visual inspection technologies and machine learning algorithms present Johnson & Johnson Vision with an opportunity to enhance quality control for Acuvue contact lenses, addressing inefficiencies such as unnecessary scrap, customer complaints, and lead time variability. With over 5 billion lenses produced annually across 100 manufacturing lines, the proposed inspection implementation of advanced camera optics and machine learning aims to improve defect detection accuracy, minimize manual inspection, and reduce customer complaints. An impact evaluation and prioritization framework was developed to strategically implement these upgrades across 100 manufacturing lines, integrating historical data analysis, financial modeling, and engineering risk assessments. Key findings highlight that complaint reduction, scrap savings, and labor cost reductions are the primary drivers of cost savings, with inventory savings offering incremental benefits over time. In conclusion, this research demonstrates the process of integrating advanced technologies into manufacturing processes. By aligning engineering solutions with strategic business objectives, the findings provide actionable insights for managing large-scale technological upgrades across global networks.

AI Through the Viewfinder: Reimagining the Camera as a Tool for AI Image Generation

Wed, 01 May 2024 00:00:00 GMT

AI Through the Viewfinder: Reimagining the Camera as a Tool for AI Image Generation Shodipo, Bukunmi The rapid emergence of artificial intelligence (AI) is causing profound shifts within the art world, reigniting age-old debates on the boundaries of what can be considered art. For example, many AI systems are employed to mimic the styles of existing artists and their works. Although this approach is deemed to be derivative and uninspiring to many people in the art world, it is also forcing us to reconsider longstanding beliefs attached to creativity such as the importance of originality and authorship. Given that AI is here to stay, this thesis explores a critical question around AI and perception, asking “How and what does AI see? Specifically, this thesis investigates the types of biases that are ingrained or embedded into AI systems, and how these biases are reflected in the output, specifically in the context of images. As part of this investigation, this thesis culminates with a prototype - an AI camera that embodies the process of AI ‘seeing the world’. This camera integrates photography with artificial intelligence, serving not only as a tool for technical exploration but also as a metaphor for examining how AI technologies offer diverse and potentially transformative perspectives on reality, much like a traditional camera. By abstracting AI technology into a camera, this project aims to start a conversation about how AI, like a camera, offers us different, sometimes biased views of the world. In doing so, the camera is redefined from a mere tool for capturing images to one that generates them, and in some cases (mis)represents human forms and identities.

The Technical Discourse of Miyake Design Studio: Episode in the Interpretation of Cloth, 1995-2007

Wed, 01 May 2024 00:00:00 GMT

The Technical Discourse of Miyake Design Studio: Episode in the Interpretation of Cloth, 1995-2007 Tan, Yi-Ern Samuel In the late 1980s, Miyake Design Studio began to register patents concerning the Studio’s development of novel techniques to process pleated clothing. Their first patent, filed in 1989, was registered in designer Issey Miyake’s name, detailing the use of an industrial machine to pleat an entire garment after sewing, reversing the order of the conventional approach to creating pleated garments. In the years that followed, this entry into what I term “technical discourse” would proliferate with the Studio’s establishing of the PLEATS PLEASE brand specializing in pleated garments, and the A-POC (“a piece of cloth”) project with designer and textile engineer Fujiwara Dai. Each of these projects produced numerous patents, including a period between 1997 and 2008 I call the “Miyake Patent Explosion” when the Studio filed twenty patents with the Japan Patent Office and its international counterparts. In contrast to aesthetic discourses proposing the value of a work on its artistic merits and intellectual content, technical discourse points to the profusion of texts produced and circulated by the Studio—in this thesis, patents and legal claims—to uphold the utility of their products and their protection as intellectual property. By engaging with technical discourse, Miyake Design Studio were not only creating legal safeguards around the ideas it considered proprietary. Rather, their extensive production of technical discourse positioned Miyake as a figure who exceeded the boundaries of fashion, approaching its adjacent categories of unhyphenated design, architecture, and art within whose circles his objects circulate as currency. Exploring these texts as they are deployed in the defense of intellectual property, I argue that technical discourse can be treated as a form of historical archive that allows us to historicize claims to technological inheritance that bear upon the discussion of Miyake’s work. Specifically, I look to patents as a citational practice, or as Alain Pottage and Brad Sherman write, a “chain of reference” through which patent lawyers and engineers make deliberate connections between one technology and another to acknowledge, distinguish, and legitimize. Examining three episodes where technical discourse opens the way for historical narrative—a lawsuit over imitation goods, a case of mistaken identity in design criticism, and a moment of technological dissolution—I argue that we cannot divorce Miyake and his work from the technical complex that surrounds the Studio’s production of objects. Turning to these technical discourses that exist in the public record, I suspend the promise of monographic history that peers into the mind of the individual and probe instead the possibilities of seeing agencies beyond those attributed to the authorial figure of Miyake— his corporate apparatus, his allies, his admirers, his critics, his opponents, the receptive public.

How Joint Inference of the Lexicon and Phonology Affects the Learnability of Process Interactions

Fri, 01 Sep 2023 00:00:00 GMT

How Joint Inference of the Lexicon and Phonology Affects the Learnability of Process Interactions Yang, Christopher Contemporary phonological research has increasingly become interested in exploring the topic of learnability through the use of computational models. However, many of the proposed models lack one or more of the following properties. (1) Many models do not consider the effect of the lexicon at all on performance, and those that do fail to consider the effect contextual allomorphy has on performance. (2) Many models characterize learnability in terms of the algorithmic implementation of search, rather than a more principled relationship between the data and the hypothesis space. These properties are critically relevant when it comes to the learnability of process interactions. The experimental literature has demonstrated that artificial languages exhibiting patterns generated from certain process interactions are more likely to be successfully reproduced and generalized by participants than others (Ettlinger 2008; Kim 2012; Brooks, Pajak, & Baković 2013; Prickett 2019). The historical literature has likewise noted that surface patterns generated from particular process interactions are more likely to change in systematic ways than others, including lexicalization, in which an alternation is encoded into the lexicon instead of the phonology, and reanalysis, in which a surface generalization is lost or changed entirely (Kiparsky 1968, 1971). Each of these hypotheses make different predictions when generating forms not seen during training. In this dissertation, I make the following contributions. (1) I propose a novel noisy-channel model of morphophonological learning. This model jointly infers a weighted space of consistent and nearly consistent lexicons and grammars from labelled, unparsed surface data. Predictions are generated given the entirety of the inferred weighted space. (2) I compare the predictions of the model to the results two artificial language learning experiments, which, despite involving the same underlying processes, produced contradictory results. I show that the model is able to achieve the results of both experiments under a unified account: the generalizability of a pattern is determined by the number of hypotheses compatible or nearly compatible with that pattern.

Investigation of consolidation and plastic resistance on clays

Sat, 01 Jan 1944 00:00:00 GMT

Investigation of consolidation and plastic resistance on clays Marsal, Raúl J. Thesis: M.S., Massachusetts Institute of Technology, Department of Civil and Sanitary Engineering, 1944; Vita. Appendix contains numerous pamphlets.

An anthropological study based upon observations of complexion and cephalic measurements of students at the Massachusetts Institute of Technology

Wed, 01 Jan 1896 00:00:00 GMT

An anthropological study based upon observations of complexion and cephalic measurements of students at the Massachusetts Institute of Technology Fisk, Harry George.; Melluish, James George. Thesis: B.S., Massachusetts Institute of Technology, Department of General Studies, 1896

The design and construction of a new density photometer

Mon, 01 Jan 1923 00:00:00 GMT

The design and construction of a new density photometer Brown, Sherwood Fiske.; Perkins, Oliver L. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrochemical Engineering, 1923; Includes bibliographical references (leaves 17-18).

An optical instrument for the synthesis of sound

Wed, 01 Jan 1930 00:00:00 GMT

An optical instrument for the synthesis of sound Brown, Sherwood Fiske. Thesis: M.S., Massachusetts Institute of Technology, Department of Physics, 1930

The chemical and physical constitution of reduced copper-red glazes

Sun, 01 Jan 1961 00:00:00 GMT

The chemical and physical constitution of reduced copper-red glazes Brown, Sherwood Fiske. Thesis: Sc. D., Massachusetts Institute of Technology, Department of Metallurgy, 1961; Includes bibliographical references (leaf 60).

The Pleasant Valley, Nova Scotia, Limestone

Thu, 01 Jan 1959 00:00:00 GMT

The Pleasant Valley, Nova Scotia, Limestone Jeffries, James T.; Manlove, Robert F. Thesis: B.S., Massachusetts Institute of Technology, Department of Geology, 1959; Includes bibliographical references (leaf 63).

A spiritual and cultural synagogue center for modern American Jewry : an architectural and sociological study

Mon, 01 Jan 1951 00:00:00 GMT

A spiritual and cultural synagogue center for modern American Jewry : an architectural and sociological study Goody, Marvin E. Thesis: M. Arch., Massachusetts Institute of Technology, Department of Architecture, 1951; "A thesis submitted in partial fulfillment of the requirements for the degree of Master in Architecture, Massachusetts Institute of Technology, August 22, 1951."; Includes bibliographical references (leaves 93-95).

Essays on structuralism and development

Thu, 01 Jan 1981 00:00:00 GMT

Essays on structuralism and development Boutros-Ghali, Y. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Economics, 1981; Includes bibliographies.

Design and evaluation of a frequency-shifting hearing aid.

Sun, 01 Jan 1978 00:00:00 GMT

Design and evaluation of a frequency-shifting hearing aid. Falkenburg, Douglas Emil. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 1978; Bibliography: leaves 103-104.

Neutron scattering study of the magnetism and structural phases of superconducting La₂CuO₄₊y̳

Sat, 01 Jan 2000 00:00:00 GMT

Neutron scattering study of the magnetism and structural phases of superconducting La₂CuO₄₊y̳ Lee, Young Sang, 1971- Thesis: Ph. D., Massachusetts Institute of Technology, Department of Physics, 2000; In title on t.p., double-underscored "y" appears as subscript.; Includes bibliographical references (p. 195-215).

An Operational Value Stream Analysis for Developmental Excellence

Thu, 01 May 2025 00:00:00 GMT

An Operational Value Stream Analysis for Developmental Excellence Shaw, Eric T. The aerospace and defense industry faces increasing challenges in new product development, where financial constraints and risk aversion hinder innovation. Using a multidisciplinary approach that integrates contract theory, computational fluid dynamics (CFD), and machine learning, this research explores the impacts of engineering requirements, financial alignment among stakeholders, and improved efficiencies in predictive modeling techniques for two separate air vehicle programs: A and B. A Monte Carlo analysis using SEER-H estimation software quantifies the financial and schedule impacts of engineering requirements, revealing a 10–30% cost increase due to volatility in air vehicle development design parameters. Moreover, a game-theoretic contract negotiation simulation illustrates the importance and opportunity of financial incentive alignment among key stakeholders. Additionally, predictive analytics leveraging machine learning models better capture the relevant flow mechanics, improving the circumferential distortion estimations in nacelle aerodynamics by over 10% compared to traditional heuristics. Finally, a CFD-based actuator disk source modeling approach demonstrates a 60% reduction in steady-state distortion at some portions of the flight envelope, due to the impact of the fan upstream influence on inlet flow distortion suggesting increased operational capability for the air vehicle program B. This research provides actionable recommendations to enhance the operational value stream of new air vehicle program development, emphasizing the need for pre-RFP requirements validation, advanced machine learning applications for predictive engineering, and refined CFD modeling to identify technical risks earlier in the design process.

A Computational Framework for Simulating Entanglement-Based Drone Countermeasures with Flexible Filaments Immersed in Viscous Flow

Thu, 01 May 2025 00:00:00 GMT

A Computational Framework for Simulating Entanglement-Based Drone Countermeasures with Flexible Filaments Immersed in Viscous Flow Sonandres, Jake T. In this work, we present a computational framework for modeling the coupled dynamic interactions of highly flexible slender filaments immersed in a viscous flow and their entanglement with themselves and moving structures. This work is motivated by a novel drone countermeasure that entangles propellers with flexible filament clouds, inducing a loss of thrust and control authority in the drone. However, the framework is relevant to a wider range of applications, including actin filaments in cell biology, carbon nanotubes in composite materials, and rope-like structures in industrial settings. Each filament is modeled with the three-dimensional geometrically exact Kirchhoff-Love torsion-free finite element beam formulation. The fluid flow resulting from filament aerodynamic interaction is described through a Boundary Integral (BI) formulation of the incompressible Stokes equations based on the Stokeslet discretization. The heavy computational load of the resulting dense system is addressed through the use of fast GPU-based dense linear solvers. The BI formulation is coupled to the filament solid mechanics by enforcing momentum balance at the dynamically evolving filament-fluid interface. Additionally, the solid contact interactions between filaments are modeled with a point-to-point frictional contact algorithm that applies discrete contact and frictional forces at the closest point between the beam elements. We address the difficulties associated with contact between elements represented with third-order Hermitian polynomial shape functions and the strategies adopted to overcome these challenges. To capture propeller fouling for drone countermeasures, we incorporate a propeller and motor model whose thrust and torque responses are affected by contact interactions during entanglement. We verify our framework against simple analytical solutions and demonstrate its capabilities with numerical examples that attempt to capture large-scale filament entanglement behavior. In particular, we apply our methodology to demonstrate the process by which filament entanglement can restrict motion and reduce the efficacy of propellers. The results show that the framework can be used to understand the connection between filament entanglement, key system properties, and the resulting thrust generated by the propeller.

Global sustainable aviation fuel production potential from current agricultural production: a holistic data analytics and systems analysis approach

Thu, 01 May 2025 00:00:00 GMT

Global sustainable aviation fuel production potential from current agricultural production: a holistic data analytics and systems analysis approach Martin, Estelle Claude Aline Aviation contributes significantly to global greenhouse gas emissions, driven primarily by its dependency on fossil-based jet fuel. Sustainable Aviation Fuel (SAF) offers a short-term option to mitigate these emissions. However, its current scalability remains limited, constrained by access to sustainable biomass. Realizing SAF’s potential in the near term, using the agricultural and industrial systems already in place requires a detailed understanding of biomass availability, resource competition, and the scalability of SAF production. This thesis presents a comprehensive system analysis framework and a data-driven methodology for evaluating SAF production potential based on current agricultural output, without assuming land expansion or major yield improvements and preserving food utilization. It evaluates the SAF production potential from increasing biomass availability by redirecting biomass currently used for some non-food purposes, and by utilizing processing and agricultural residue. In-depth analysis of four high-potential case studies, one for each main biomass family (starchy, sugary, oily, and fats and greases), was used to construct a detailed model of the supply chain. This structure was then applied globally across all countries and relevant feedstocks to estimate SAF production potential and associated system requirements. Findings from the case studies show that these four high-potential opportunities could collectively meet only up to 13.1% of global jet fuel demand in 2023, assuming 100% neat SAF. The global analysis estimates that the SAF production potential from the considered streams of increased biomass availability could meet up to about two-thirds of global jet fuel demand, with 28.7% derived from agricultural residues, 25.9% from redirected main products, and 12.5% from processing residues. These contributions hence remain insufficient to fully displace fossil jet fuel. This work provides an estimate of what could be achieved using the existing agricultural and industrial systems, what resource would be required, and how it compares to global resource availability.

Causal inference for complex systems and applications to turbulent flows

Thu, 01 May 2025 00:00:00 GMT

Causal inference for complex systems and applications to turbulent flows Sánchez, Álvaro Martínez Causality lies at the heart of scientific inquiry, serving as the fundamental basis for understanding interactions among variables in physical systems. Despite its central role, current methods for causal inference face significant challenges due to nonlinear dependencies, stochastic interactions, self-causation, collider effects, and influences from exogenous factors, among others. While existing methods can effectively address some of these challenges, no single approach has successfully integrated all these aspects. Here, we address these challenges with SURD: Synergistic-Unique-Redundant Decomposition of causality (Nat. Commun., vol. 15, 2024, p. 9296). SURD quantifies causality as the increments of redundant, unique, and synergistic information gained about future events from past observations. The formulation is non-intrusive and applicable to both computational and experimental investigations, even when samples are scarce. We benchmark SURD in scenarios that pose significant challenges for causal inference and demonstrate that it offers a more reliable quantification of causality compared to previous methods. We further illustrate the applicability of our approach in two turbulent-flow scenarios: the energy transfer across scales in the turbulent energy cascade and the interaction between motions across scales in a turbulent boundary layer. Our results show that, without accounting for redundant and synergistic effects, traditional approaches to causal inference may lead to incomplete or misleading conclusions.

Systems Theoretic Process Analysis of Sociotechnical Systems

Thu, 01 May 2025 00:00:00 GMT

Systems Theoretic Process Analysis of Sociotechnical Systems Harrington, Polly The safety and success of complex modern systems, such as hospitals, aircraft, or software, depend on their ability to integrate people and technical components. For example, doctors must be able to use their computerized surgical tools to treat their patients successfully, airplane pilots must be able to operate the required controls for takeoff and landing, and regulators must be able to interpret the data they receive to make critical decisions. However, designing systems that facilitate safe interactions between humans and technology is not a simple task. System designers must consider not only the constraints of the technical components but also human requirements throughout the entire system. However, accidents in modern systems continue to prove that more work is needed to identify and prevent unsafe interactions between humans and technology Systems Theoretic Process Analysis (STPA) is a hazard analysis methodology based on systems theory that has been used to improve system safety in various industries, including healthcare, aviation, nuclear power, and automotive design. However, if hazard analysts using STPA lack significant expertise in human factors engineering (HFE), they may be unable to thoroughly and rigorously identify critical unsafe interactions. This thesis presents a process for utilizing HFE to improve the results of STPA analyses on sociotechnical systems. In particular, the process focuses on the thorough identification of causal scenarios in sociotechnical systems by incorporating relevant human factors concepts. The process allows analysts without significant training in HFE to improve their ability to identify useful scenarios for humans in their system. The effectiveness of the improved process is demonstrated using a healthcare case study on over-the-counter clinical laboratory tests in the United States. By establishing a process for non-HFE experts to use when conducting STPA analyses, more systems can be developed that enhance human performance rather than increase conflict between humans and the engineered system.

Development of Algorithms for Quantitative Analysis of Long Electrical Arcs in Crossflows

Thu, 01 May 2025 00:00:00 GMT

Development of Algorithms for Quantitative Analysis of Long Electrical Arcs in Crossflows Lin, Fayleon A single lightning strike can deliver a steady current of hundreds of amps during its attachment to an aircraft. Therefore, it is imperative to have an adequate lightning protection system in the aircraft to minimize the probability of catastrophic accidents. Current guidelines for lightning protection systems are based on prior service experience and historical data, which might become insufficient with future generation aircraft. These often adopt novel and unconventional aircraft designs, often deviating significantly from current designs. Therefore, efforts are underway to update these guidelines with novel methods such as designs aided by numerical simulation that can accurately model the behavior of lightning attachment and the subsequent swept-stroke phase. To aid in the development of these numerical methods, ample data of not only the electrical arcs but also their interactions with the surrounding flow are necessary for validation. However, most studies on long electrical arcs lack a detailed investigation of the coupling between the electrical arcs and the surrounding flow field. For that purpose, teams from the Massachusetts Institute of Technology (MIT), ONERA, and Universitat Politècnica de Catalunya (UPC) conducted an extensive experimental campaign in April 2024 that investigates this coupling in detail for the first time. Data gathered from this experiment include electrical properties of the arc, high-speed video of the arc column, and the velocity field of the surrounding flow. Approximately 200 cases were conducted with various geometrical and electrical configurations. To meaningfully analyze all the data, a set of algorithms was developed to automatically process, analyze, and visualize these data. Detailed analysis of the root and column behavior was performed; electrical properties were verified to be consistent with literature values; and coupling between the velocities of the arc column and the flow field was determined by simultaneous visualization of both data forms.

Performance and Analysis of a Deployable Diffractive Optical Element for Small Satellite Missions

Thu, 01 May 2025 00:00:00 GMT

Performance and Analysis of a Deployable Diffractive Optical Element for Small Satellite Missions Bahlous-Boldi, Adam A. As space missions push toward smaller, lighter, and more deployable instrumentation, diffractive optical elements (DOEs) offer a compelling alternative to traditional optics. Their ability to focus light through engineered phase profiles rather than curved surfaces allows for large-aperture, flat optics that are far lighter and easier to package for launch. However, this benefit comes with trade-offs: DOEs are sensitive to wavelength mismatch, manufacturing errors, and environmental deformations—especially thermal gradients and membrane tensioning in space. This thesis develops a comprehensive framework for understanding and simulating the performance of DOEs under realistic operating conditions. Beginning from first principles, the work contrasts geometric and wave-optical models for Fresnel zone plates and multilevel diffractive lenses, leading to quantitative predictions of diffraction efficiency and PSF quality under non-idealities. A key contribution is the analytical and numerical analysis of how uniform thickness errors, wavelength mismatches, and thermal expansions degrade optical performance, both in efficiency and wavefront fidelity. To evaluate these effects in detail, a flexible simulation tool was developed in MATLAB, enabling both Fourier and integral-based propagation through arbitrarily deformed DOEs. These models are applied to a conceptual space-based LIDAR system—SPECIES—that uses a deployable DOE optic to demonstrate the feasibility and limitations of this approach. The results show that DOEs can tolerate some global deformations - for example, a 1 mm deformation results in a 38% performance loss in an F3 LiDAR system with a 1 mm detector diameter. However, they remain highly sensitive to fine-scale shape errors, posing significant challenges for high-precision applications like fiber coupling or imaging. The findings provide new insight into the tolerances, benefits, and trade-offs of DOEbased systems in space, and lay the groundwork for future missions seeking to leverage lightweight diffractive optics for remote sensing and optical communication.

Partial Gravity Load Simulation Using Mechanical Off-Loading and Lower Body Negative Pressure

Thu, 01 May 2025 00:00:00 GMT

Partial Gravity Load Simulation Using Mechanical Off-Loading and Lower Body Negative Pressure Davalos, Daniela L. Prolonged exposure to reduced gravity environments can lead to significant deconditioning of the cardiovascular, musculoskeletal, and ocular systems. These effects increase the risk of orthostatic intolerance, bone loss, and conditions such as Spaceflight Associated Neuro-ocular Syndrome (SANS). As spaceflight missions grow longer and more frequent, especially with increased extravehicular activity (EVA) on the Moon or Mars, it is critical to develop effective countermeasures and Earth-based analogs to simulate these gravitational environments and evaluate physiological impacts. This thesis addresses these challenges through two complementary approaches. First, it presents the design and development of the MIT Moonwalker IV, a passive mechanical offloading system that simulates partial gravity by applying vertical support via a spring-cable mechanism. In a treadmill-based pilot study, one participant showed at least a 50% reduction in metabolic demand while running under simulated Martian gravity. These findings validate the Moonwalker IV as a metabolic analog for EVA task simulation. Second, this thesis evaluates a collapsible lower body negative pressure (LBNP) suit as a wearable countermeasure for micro and partial gravity environments. By applying negative pressure to the lower body, the suit helps restore the mechanical loading and hydrostatic fluid gradients typically provided by Earth’s gravity. The suit was tested in both simulated reduced gravity via a head-down/head-up tilt paradigm and and true reduced gravity via parabolic flight. Each condition was evaluated both with and without –20 mmHg of LBNP. Results demonstrated that the collapsible LBNP suit produced cardiovascular responses comparable to those observed in traditional rigid LBNP chambers. It also induced lower body fluid shifts as measured by segmental leg bioimpedance, reduced intraocular pressure, and generated ground reaction forces similar to standing in 1G. These findings support the complementary use of Earth-based analog systems to simulate partial gravity and wearable devices to simulate Earth gravity in reduced gravity environments. They offer valuable tools for preparing astronauts and preserving physiological health during long-duration space missions.

Unforgettable Generalization in Language Models

Thu, 01 May 2025 00:00:00 GMT

Unforgettable Generalization in Language Models Zhang, Eric When language models (LMs) are trained to forget (or “unlearn”) a skill, how precisely does their behavior change? We study the behavior of transformer LMs in which tasks have been forgotten via fine-tuning on randomized labels. Such LMs learn to generate near-random predictions for individual examples in the “training” set used for forgetting. Across tasks, however, LMs exhibit extreme variability in whether LM predictions change on examples outside the training set. In some tasks (like entailment classification), forgetting generalizes robustly, and causes models to produce uninformative predictions on new task instances; in other tasks (like physical commonsense reasoning and scientific question answering) forgetting affects only the training examples, and models continue to perform the “forgotten” task accurately even for examples very similar to those that appeared in the training set. Dataset difficulty is not predictive of whether a behavior can be forgotten; instead, generalization in forgetting is (weakly) predicted by the confidence of LMs’ initial task predictions and the variability of LM representations of training data, with low confidence and low variability both associated with greater generalization. Perhaps most surprisingly, random-label forgetting appears to be somewhat insensitive to the contents of the training set: for example, models trained on science questions with random labels continue to answer other science questions accurately, but begin to produce random labels on entailment classification tasks. Finally, we show that even generalizable forgetting is shallow: linear probes trained on LMs’ representations can still perform tasks reliably after forgetting. Our results highlight the difficulty and unpredictability of performing targeted skill removal from models via fine-tuning.

Guessing Random Additive Noise Decoding in Coded Multiple-Input Multiple-Output Systems

Thu, 01 May 2025 00:00:00 GMT

Guessing Random Additive Noise Decoding in Coded Multiple-Input Multiple-Output Systems Wu, Benjamin Multiple-Input Multiple-Output (MIMO) wireless communication systems incorporate forward error correction (FEC) to achieve high reliability under fading and interference. In this thesis, we explore the emerging FEC paradigm of Guessing Random Additive Noise Decoding (GRAND) in a point-to-point MIMO system. Treating GRAND as an FEC decoder disjoint from the MIMO detector, we compare the soft-decision Ordered Reliability Bits GRAND (ORBGRAND) to CRC-Assisted Successive Cancellation List (CA-SCL) decoding of the CRC-Assisted Polar (CA-Polar) [105, 128] code found in the 5G New Radio standard. For this code, we find that ORBGRAND outperforms CA-SCL (list size 16) by 1 dB E_b/N₀ at block error rate of 10⁻³, under 16-QAM and Linear Minimum Mean Square Error detection, with two transmit antennas and four receive antennas. We also show that ORBGRAND, when paired with other moderate redundancy linear codes, can yield substantial savings in the range of 0.5 − 2 dB in E_b/N₀ over CA-SCL decoding (list size 16) of CA-Polar codes with the same code parameters, for a block error rate of 10⁻³. We provide extensive benchmarks comparing ORBGRAND to CA-SCL and other soft-decision GRAND variants. We also integrate a GRAND decoder producing soft output into a MIMO iterative detection and decoding (IDD) receiver. Specifically, we apply an established technique which utilizes soft-output GRAND as the component decoder for the block turbo decoding of product codes. This block turbo decoder is evaluated as a soft output decoder within a MIMO IDD receiver. We demonstrate competitive or superior performance relative to Belief Propagation (BP) decoding of 5G Low-Density Parity Check (LDPC) codes. This approach also marks a use of GRAND for low-rate, high-redundancy FEC in a MIMO system. With GRAND in MIMO still being an emerging area of research, this work is an exploratory evaluation of GRAND for FEC in MIMO, and highlights GRAND’s potential as a versatile and performant decoder in different MIMO receiver architectures.

Improving Accuracy Predictions of Companion Classifiers for LLM Routing

Thu, 01 May 2025 00:00:00 GMT

Improving Accuracy Predictions of Companion Classifiers for LLM Routing Wu, Jessica L. The increasing versatility of Large Language Models (LLMs) calls for developing effective routing systems to match tasks with the most suitable models, balancing accuracy and computational cost. This research introduces a novel meta-cascade routing framework that combines meta-routing, where a predictive model selects the appropriate LLM for a task, and cascading, where models are queried in sequence to optimize cost and performance. A critical component of this framework is the companion classifier, defined as a fine-tuned model trained to predict whether a particular LLM will generate an accurate response. We investigate whether incorporating features such as model responses into these classifiers can improve routing accuracy. Our preliminary experiments, using the Routerbench dataset, focus on training companion models that provide more stable and accurate routing decisions.

Formal Verification of Relational Algebra Transformations in Fiat2 Using Coq

Thu, 01 May 2025 00:00:00 GMT

Formal Verification of Relational Algebra Transformations in Fiat2 Using Coq Teshome, Christian Data-intensive applications often involve operations over structured datasets, such as filtering, joining, and projecting records. Relational database systems generally use query planners to optimize high-level SQL queries into efficient execution plans. While these systems apply well-established query transformations, they typically assume the correctness of these transformations rather than formally proving them. The absence of formal guarantees can be a significant limitation for systems with strict correctness requirements. This thesis contributes to Fiat2, a Python-like high-level programming language for data-intensive workloads that integrates formal verification via the Coq proof assistant. We focus on proving the correctness of several rewrite-based query optimizations commonly used in database engines. Specifically, we formalize and prove the correctness of algebraic rewrites involving combinations of filters, joins, and projections, as well as join-reordering rewrites. All rewrites are proven in Coq to preserve the semantics of the original program under list semantics, meaning that the output lists are fully equivalent (or permutations, in the case of join reordering). These verified rewrites serve as a foundation for future optimization in Fiat2, enabling significant optimizations while preserving the semantics of the original queries with correctness guarantees. The results demonstrate the feasibility of integrating formally verified query optimizations into a practical high-level programming language.

Converting PyTorch Models to StreamIt Pipelines

Thu, 01 May 2025 00:00:00 GMT

Converting PyTorch Models to StreamIt Pipelines Rajvee, Muhender Raj With the rise of large language models, there have been efforts to optimize machine learning inference to support a large volume of queries. Currently, the two main ways to do this are running optimized kernels for computing the forward inference pass and distributing computation across multiple GPUs or different cores in a GPU. Machine learning libraries such as PyTorch produce dynamic computation graphs in order to represent the forward pass of the model. PyTorch allows conversion of these dynamic graphs into static ones through just-in-time (JIT) compilation. These graphs can then be optimized further by the compiler. We propose an alternate way of optimizing these dynamic graphs. We convert the dynamic computation graph of PyTorch to pipelines in StreamIt, a domain specific language (DSL) for streaming applications, and use the multi-stage compilation property of BuildIt to compile this pipeline in stages to inference code. We found that, while the inference latencies of models compiled in this way are slightly higher, they are still comparable to those of PyTorch models and are open to future optimizations.

An Interactive Visual Paradigm for Knowledge Graph Question-Answering

Thu, 01 May 2025 00:00:00 GMT

An Interactive Visual Paradigm for Knowledge Graph Question-Answering Ramkumar, Vayd Sai In an era of information overload, verifying data reliability and provenance is critical, yet knowledge graphs (KGs) often remain complex for non-expert users. This thesis introduces TRACE, a Reasoning and Answer-path Comprehension Engine, a visualization tool enhancing transparency in KG question answering (KGQA). By abstracting intricate KGs into intuitive meta-nodes, TRACE simplifies exploration of large, multi-topic datasets. Its interactive interface allows users to navigate semantic communities and trace reasoning paths, fostering trust through clear answer derivation. Unlike cluttered traditional graph visualizations, TRACE’s meta-node approach provides a scalable, user-friendly solution, concealing technical complexities while enabling robust query validation. Large language models support natural language query parsing and community summarization, making KGs accessible to diverse audiences. TRACE positions itself as a vital widget for information platforms, empowering users to counter misinformation confidently. A user study and pipeline evaluation confirmed TRACE’s intuitive interface excels for complex queries, though multi-hop paths pose challenges, while processing tests demonstrated its scalable paradigm for large datasets. By prioritizing transparency and usability, TRACE redefines KGs as reliable tools for knowledge discovery, laying a foundation for future systems to deliver trustworthy, accessible information in a digital landscape fraught with uncertainty.

Spectral Analysis of Local Atomic Environments

Thu, 01 May 2025 00:00:00 GMT

Spectral Analysis of Local Atomic Environments Phung, Tuong The representation of local environments is a cornerstone challenge in computational materials science, with profound implications for property prediction and materials discovery. This thesis presents a comprehensive investigation of spectral descriptors constructed from spherical harmonic expansions to represent the geometries of local atomic environments. Systematic computational experiments evaluate the robustness of these descriptors to geometric perturbations and their capacity to differentiate structurally similar configurations. The findings reveal a clear performance hierarchy, with higher-order descriptors offering increased geometric expressivity and reconstruction accuracy in resolving challenging structural cases. This research further examines methods for inverting spectral representations back to atomic coordinates, demonstrating that directly optimizing three-dimensional positions through gradient-based techniques yields markedly better reconstruction accuracy than approaches operating in Fourier space. Dimensionality reduction via latent space embeddings is also explored, showing that essential geometric features can be preserved in significantly compressed representations. Through methodical analysis of descriptor limitations, performance boundaries, and sensitivity to hyperparameters, this work establishes practical benchmarks and implementation guidelines for spectral descriptors. These contributions strengthen the foundation for reliable machine learning models in computational materials science, advancing both the accuracy and efficiency of atomic-scale modeling for materials design and discovery.

Design and Optimization of Shipping Container for Package-Less Units

Thu, 01 May 2025 00:00:00 GMT

Design and Optimization of Shipping Container for Package-Less Units Minja, Baraka Package-less shipping aims to deliver units without company X’s added packaging. This requires the fulfillment systems and processes to have gentler handling. Part of this change involves the design and implementation of a container that will carry units from a distribution center to a delivery facility. This thesis presents the container analysis that was completed to determine what the optimal container features and container type are for package-less shipping. Collapsible bags provide the best solution for package-less shipping in comparison to nestable and collapsible totes. Since ergonomic weight is the limiting constraint, the lower weight of the collapsible bag will allow for 1 or 2 more units per container. In addition, it benefits from 1) lower process cost for returning to dock (3.7% cost reduction as compared to a nestable tote) 2) better ergonomics (collapsible tote has undesirable pinch points) and 3) improved cycle time (estimated 2s to open/collapse compared to 4s for collapsible tote). Additional considerations that require more analysis relate to units per container and relocation. Based on company X’s past orders and unit types for the package-less shipping process, it is estimated that ~210 units per container (17.08 cu. Ft.) is the max achievable for NA before it reaches the ergonomic weight cap. However, company X is expecting the package-less shipping distribution center process to be constrained to ~105-133 units. Analysis of container relocation from delivery facilities to distribution centers indicates it is worthwhile investigating alternative relocation strategies in lieu of dedicated 53-foot container trailers to achieve lower relocation costs. The collapsible bag is the best option assuming it has at least an expected lifetime of 2 years, which is when its NPV exceeds that of the two alternatives. These results are sensitive to assumptions made, and it is necessary to fine tune this analysis when the end-to-end package-less shipping process has been fully mapped out.

Transformation Tolerance of Facial Recognition Technology and Informative Evaluation Metrics

Thu, 01 May 2025 00:00:00 GMT

Transformation Tolerance of Facial Recognition Technology and Informative Evaluation Metrics Nakamura, Haley Marie Over the last decade, machine learning based facial recognition (FR) systems have continued to increase in popularity while spreading to unique deployment settings. Despite the large variance among FR input distributions, popular facial recognition benchmarks continue to characterize system performance using one aggregate score over a single dataset. In many cases, the limitations of this score are unclear to downstream users: assuming benchmark accuracy is high, how is it expected to change for an image sampled from a distinct distribution? Which transformations can the model handle robustly, and which cause failure? Meanwhile, there is a large body of human facial perception research that aims to understand the underlying mechanisms of human recognition. This field offers methodological inspiration for more informative evaluation techniques, including the characterization of recognition performance as a function of a quantifiable input transformation. This work performs such an analysis. The performance scores of five state-of-the-art FR models are characterized as a function of Gaussian blur strength, intersecting with color variation. The performance-blur relationship is modeled as an s-curve, creating a highly interpretable format for discussion. Blur strength was consistently statistically significant to performance, but color variation did not significantly impact any model. Results are then compared to prior human recognition experiments. The best models outperform humans in low-blur regimes while humans outperform all models in high-blur regimes. These results motivate the need for modern benchmarks that capture a range of input distributions. The analysis presented can lead to a deeper understanding of FR systems, and provide a clearer interpretation of how model performance changes under quantified distribution shifts.

Design Transfer as a Lever for Accelerated Medical Device Innovation: A Case-Based Mapping Approach

Thu, 01 May 2025 00:00:00 GMT

Design Transfer as a Lever for Accelerated Medical Device Innovation: A Case-Based Mapping Approach Magzoub, Amna Ahmed Eltayeb In highly regulated industries such as medical devices, accelerating New Product Development (NPD) without compromising quality or compliance is a persistent challenge. This thesis investigates the design transfer process, a critical, yet under- examined phase of NPD, as a strategic lever to reduce time-to-market. The project uses swimlane flowcharts and Design Structure Matrices (DSM) to map real-world processes, identify breakpoints, and classify rework (both planned and unplanned) in four case studies from Stryker Corporation. Key patterns emerged across case types: insufficient early-stage validation, misaligned cross-functional communication, and inadequate integration with suppliers were recurrent drivers of inefficiency. Compara- tive analysis revealed that concurrent engineering practices and knowledge sharing significantly reduce unplanned rework cycles and improve development speed. The study proposes actionable recommendations for optimizing design transfer including: leveraging corporate know-how through intentional knowledge transfer meetings dur- ing the process benchmarking process, increased risk-taking during the development process by embracing concurrent engineering approaches, and investing in early-stage co-development by adopting regular collaboration activities with suppliers. These findings can inform broader process improvements in the development of medical devices, and serve as a blueprint for other complex, cross-functional environments.

The Effect of the Solar Cycle on Satellite Orbital Lifetime

Thu, 01 May 2025 00:00:00 GMT

The Effect of the Solar Cycle on Satellite Orbital Lifetime Lisy, Celvi A. The lifetime of a satellite in Low Earth Orbit (LEO) is affected by the 11-year solar cycle. At a fixed altitude, increasing solar activity increases atmospheric density which leads to an increase in drag, and a decrease in mission lifetime without using propulsion to recover altitude. Satellites may have longer orbital lifetimes if more of their mission is operational during a solar minimum due to lower solar activity and lower atmospheric drag. Satellites with larger area-to-mass ratios generally have shorter orbital lifetimes than satellites with small area-to-mass ratios. Missions that get delayed and have more of their operations during solar maximum than planned originally may have too short of a mission lifetime or, conversely, may be at risk of increasing their orbital lifetime past regulatory limits (five years for satellites in LEO according to the FCC) if they launch closer to solar minimum. For example, a satellite with an area-to-mass ratio of 0.014 m2/kg – such as a 1U CubeSat – and a one-year mission that is launched in 2021 without onboard propulsion, would have an orbital lifetime of 1.051 years. However, if that mission were delayed a year, a common occurrence in the industry, it would no longer be able to achieve its mission as its orbital lifetime with a deployment in 2022 is 0.44 years. Conversely, if the same 1U CubeSat is launched during solar max in January 2025, it would have an orbital lifetime of 2.2 years, and would re-enter in February of 2027. However, if that mission were delayed a year, the satellite would launch in January 2026 and instead be in orbit for 6.4 years before re-entering. They could be fined for violating the FCC deorbit limit of five years. This thesis quantifies the effect of launch or processing delays on satellite orbital lifetime based on their orbit altitude and vehicle parameters such as mass, cross sectional area, altitude, and bus size. In general, it is found that four-year and six-year delays have the greatest effect on a satellite’s orbital lifetime because the satellite will be deorbiting almost half a solar cycle (5.5 years) from its intended deployment year. However, two-year delays can still affect satellite operators, as they can increase the orbital lifetime, even by up to 1.5 years for low area-to-mass ratio satellites in 400 km orbits and almost five years for satellites in orbits higher than 500 km. Two-year delays can also decrease the orbital lifetime of a satellite by up to 1.7 years for low area-to-mass ratio satellites in 400 km orbits and almost two years at altitudes higher than 500 km.

Electrical Diagnostics for Nanosecond Pulsed Discharge Reactors

Thu, 01 May 2025 00:00:00 GMT

Electrical Diagnostics for Nanosecond Pulsed Discharge Reactors Rao, Sankarsh R. This thesis provides an introduction to transmission line theory (telegrapher’s equations) as the mathematical background needed to correctly perform and interpret electrical measurements in nanosecond pulsed discharge reactors. The mathematical framework is implemented in a numerical tool called VI-View, which is made available to the community to aid with the interpretation of electrical measurements and help explain discrepancies between different experimental arrangements and probe configurations. A brief manual on how to use the tool is provided, followed by a series of six case studies relevant to experimental setups/situations encountered in practice. The analysis of these case studies summarizes best practices when performing electrical and energy measurements in nanosecond pulsed discharge reactors. Case Studies 1 and 2 cover in-situ and remote measurements for reactors using one voltage and one current probe. Case Study 3 covers how two current probes, one on the high-voltage end and one on the low-voltage end, can achieve the same energy measurements as Case Studies 1 and 2. Case Studies 4 and 5 show how cables with varying lengths and dissimilar properties — as can sometimes be encountered in practice — affect the electrical signals. Case Study 6 shows how a variable resistance — a step drop from 50MΩ to 10Ω — within a load can be a first approximation to a plasma reactor with a discharge. Finally, an outlook on how these case studies connect to real, experimental waveforms is presented along with the limitations of the tool.

Stochastic Methods for Setting Effective Aviation NOₓ Policies

Thu, 01 May 2025 00:00:00 GMT

Stochastic Methods for Setting Effective Aviation NOₓ Policies Reider, Sarah Nitrogen Oxides (NOₓ) from aviation emissions are well known to have detrimental effects on air quality and the climate. Presently, they are regulated to preserve local air quality around airports. As part of the regulation process, aircraft engines are placed on a test stand with NOₓ levels measured at different thrust settings meant to mimic the aircraft’s emissions during landing and take-off. These are then constrained as a function of the engine’s overall pressure ratio (OPR) and rated thrust, with the allowed NOₓ emissions increasing with OPR. Despite increases in the stringency of this regulation, recent research suggests this regulation is insufficient for protecting surface air quality degradation from NOₓ emissions at cruise. Moreover, at high OPRs, NOₓ emissions increase substantially for relatively small reductions in fuel burn. In light of this, a new metric representative of cruise emissions is being investigated. This work considers effective methods to define this new regulation given a wide range of uncertainties in the tradeoff between NOₓ and CO₂ emissions at high OPRs. First, an estimate for the combined climate and air quality cost of NOₓ from aviation cruise emissions is estimated as ∼$95,000/tonne using a 2019 flight inventory. Then, cruise limits are proposed informed by the combined impact of NOₓ and CO₂ at cruise and with a similar slope to the current LTO standard. Finally, a Monte Carlo simulation is run, sampling NOₓ and CO₂ social costs for a series of hypothetical aircraft designed using the open-source Transportation Aircraft System OPTimization (TASOPT) model. This work takes a worst-case scenario approach, where the only response engine manufacturers can make to stricter standards is to reduce OPR and sacrifice fuel efficiency. Each aircraft’s emissions are evaluated during cruise to determine the probability of increasing environmental harm under different policy scenarios given these uncertainties. The combined cost of NOₓ and CO₂ are compared to the baseline engines that meet current regulations for each scenario. Results show defining a cruise metric informed by the weighted combined cost of CO₂ and NOₓ could reduce total environmental cost at cruise by 15 – 43% while carrying a 6 – 7.4% risk of increasing total environmental cost for wide-body aircraft engines in the most stringent scenario. Less stringent scenarios showed similar risks of increasing harm for smaller potential environmental savings. In all cases, the risks associated with the proposed limits are driven by low-likelihood extremes in the uncertainty distributions of NOₓ and CO₂, further suggesting the benefit of an environmentally conscious standard.

Domain-Independent Mode Estimation for Human-Robot Collaboration

Thu, 01 May 2025 00:00:00 GMT

Domain-Independent Mode Estimation for Human-Robot Collaboration Gomez, Annabel Reyna To collaborate safely and intelligently with humans, robots must infer high-level semantic sates, such as intentions or interaction modes, from uncertain sensor input. While dynamic, probabilistic mode estimation is commonly used in fault diagnosis, this thesis extends the problem to activity recognition, where the goal is to estimate qualitative, symbolic human-object interaction states in real time. Robust human activity recognition is essential for collaborative and assistive robotics, particularly in dynamic or safety-critical environments. The core solution presented in this thesis is a mode-estimator and its efficient implementation using the A* with bounding conflicts (A*BC) algorithm. This performs best-first enumeration over symbolic activity states while integrating recursive Bayesian filtering to maintain belief under noisy observations. Unlike low-level trajectory tracking or deep-learned classifiers, qualitative spatial filtering operates at the right level of abstraction to recognize symbolic actions. It can also generalize across domains with minimal retraining and support efficient, probabilistically grounded reasoning about uncertainty in both perception and symbolic mode transitions. The proposed system fuses RGB-D perception, object segmentation, qualitative spatial reasoning (QSR), and probabilistic inference into a real-time pipeline capable of tracking and inferring symbolic human-object interaction states. Evaluated in a human-robot rehabilitation setting, this domain-independent system successfully infers latent human and object activity states from noise RGB-D data. It resolves ambiguity using Vision-Language Model (VLM)-guided semantic arbitration and demonstrates robustness and adaptability in unstructured environments. This work establishes qualitative spatial filtering with A*BC as a generalizable and efficient solution for semantic activity recognition, laying the foundation for future perception-driven collaborative systems.

Distributed Energy Dynamics Control for Stable Power Electronic-Enabled Electric Power Systems

Thu, 01 May 2025 00:00:00 GMT

Distributed Energy Dynamics Control for Stable Power Electronic-Enabled Electric Power Systems Gada, Hiya Akhil The increasing penetration of renewable and inverter-based resources is transforming modern power systems into fast, nonlinear, and heterogeneous networks. These converterdominated systems operate on timescales much faster than traditional synchronous machines, making conventional modeling and control approaches, rooted in quasi-static phasor analysis and centralized architectures, inadequate for ensuring stability and scalability. This thesis adopts an energy space modeling approach grounded in first principles of energy conservation and system interconnection. It extends the previously introduced second-order energy dynamics model by relaxing the assumption that energy in tangent space can be treated as an independent disturbance. The resulting contribution is a third-order model that treats stored energy in tangent space as a dynamic state, enabling more expressive and accurate modeling of fast-timescale system behavior. Leveraging this extended energy space model, the thesis develops a multilayered distributed control architecture in which the nonlinear physical dynamics of each component are lifted to the higher-level linear energy space, capturing internal energy dynamics and real/reactive power flows, and integrated with the lower-level physical dynamics with well-defined mappings. Distributed controllers are designed in this energy space using only local states and minimal neighbor interaction, assuming a system-level coordination mechanism provides consistent references. Two control strategies, energy-based feedback linearizing control and sliding mode control, are developed and shown to achieve asymptotic convergence to reference outputs. The framework is validated on two systems: an inverter-controlled RLC circuit and a synchronous generator under load. Finally, the energy space framework is extended to structurally model inter-area oscillations (IAOs). An inter-area variable is defined as the difference between power incident on a tie-line from Area I and power reflected into tie-line from Area II. Simulations on a 3-bus, 2-area system confirm consistency with eigenmode analysis and show how tie-line strength and generator inertia affect IAO dynamics. A novel resonance phenomenon is also identified: instability arising from interaction between a system’s natural IAO frequency and time-varying disturbances from intermittent DERs. This previously unmodeled behavior is captured explicitly within the energy dynamics framework and may help explain recent blackout events in the Iberian Peninsula.

Pushing the Limits of Active Data Selection with Gradient Matching

Thu, 01 May 2025 00:00:00 GMT

Pushing the Limits of Active Data Selection with Gradient Matching Zhang, Chris As modern machine learning systems grow in scale, the inefficiencies of training on large, noisy, and imbalanced datasets have become increasingly pronounced—particularly in computer vision, where real-world data often contain labeling errors, occlusions, and redundancy. While large models can partially compensate by training exhaustively on massive datasets, this indiscriminate approach is computationally expensive and often inefficient. Active data selection offers a more efficient alternative by prioritizing examples that contribute most to model improvement. However, existing selection strategies (such as Rho Loss) still fall short of the optimal achievable performance. In this work, we propose the Gradient Informed Selection Technique (GIST), an active data selection method that prioritizes examples based on their gradient alignment with a small, fixed holdout set. At each training step, GIST computes perexample gradients and selects those that are most aligned with the holdout gradient, thereby guiding model updates toward better generalization. We evaluate GIST on noisy (Clothing1M) and clean (ImageNet) datasets and show that it consistently outperforms baselines across a range of selection ratios—that is, the proportion of a batch of data that the model selects to update weights on. To address the computational overhead of gradient-based selection, we introduce efficient variants using restricted-layer gradients, low-rank approximations, and gradient quantization. We also analyze GIST’s selection behavior, showing that it implicitly balances classes and repeatedly selects high-utility examples—two factors that enhance both robustness and learning efficiency. Our findings suggest that a more effective data curriculum is both discoverable and practical, and that GIST is a step toward achieving it.

The Phase Transition for Recovering a Random Hypergraph from its Edge Data

Thu, 01 May 2025 00:00:00 GMT

The Phase Transition for Recovering a Random Hypergraph from its Edge Data Yao, Andrew The weighted projection of a hypergraph is the weighted undirected graph with the same vertex set and edge weight equal to the number of hyperedges that contain the edge; the projection is the unweighted graph with the same vertex set and edge set consisting of edges with weight at least one. For d ≥ 3, after observing the unweighted and weighted projection of a random d-uniform hypergraph that is sampled using a generalization of the Erdős–Rényi random model, we study the recovery of a fraction of the hyperedges and the entire hypergraph. For both cases, we show that there is a sharp phase transition in the feasibility of recovery based on the density of the hypergraph, with recovery possible only when the hypergraph is sufficiently sparse. Particularly, we resolve numerous conjectures from [5]. Furthermore, we display an efficient algorithm that is optimal for both exact and partial recovery. We also analyze the phase transition for exact recovery by exhibiting a regime of probabilities that is below the exact recovery threshold by a polylogarithmic factor for which exact recovery is possible.

Empowering Mobile-Only App Generation — Offline AI Code Generation with App Inventor

Thu, 01 May 2025 00:00:00 GMT

Empowering Mobile-Only App Generation — Offline AI Code Generation with App Inventor Yuan, Joyce As digital tools become more accessible, creating software is becoming a powerful way for anyone to make real-world impact. Computational action—the idea that learners can build computing artifacts with authentic relevance to their lives and communities—reframes computing as a tool for empowerment. Low-code platforms like MIT App Inventor support this vision by fostering digital agency through purposeful creation. Recent advances in large language models (LLMs) expand these possibilities further by enabling code generation from natural language, offering a timely opportunity to lower the barrier to app creation. MIT App Inventor has long championed accessibility, allowing even young learners in underserved regions to build meaningful mobile apps. Its natural language tool, Aptly, enables users to describe app ideas and generate functional code. However, Aptly’s reliance on cloud-based LLMs limits access for users without stable internet—often those who could benefit most. This thesis addresses that challenge by enabling AI-powered app creation to run entirely offline on mobile devices. We fine-tune and quantize LLaMA 3B using QLoRA and deploy it on iOS with MLC LLM, enabling on-device inference without internet. We also introduce a custom evaluation framework tailored to Aptly’s grammar, combining a Tree-sitter parser and a modified CodeBLEU metric to assess both semantic and syntactic quality. Using curated evaluation datasets, we benchmark out-of-box and fine-tuned models across prompting strategies. In our evaluations, fine-tuned GPT-4.1 achieved the highest normalized CodeBLEU score (0.36 ± 0.12) and parsed over 81% of completions, outperforming its baseline by more than 5%. QLoRA-finetuned LLaMA improved parseability by 11.7% over its base model, showing progress in adapting smaller models to the Aptly domain, though semantic fidelity remains a challenge. Our results show that offline natural language–to–app generation is feasible, and that smaller models can be adapted to the Aptly domain. By lowering the technical and infrastructural barriers to app creation, this work lays the foundation to empower AI-assisted programming that is accessible, offline, and on the phone.

AutoDiff: A Scalable Framework for Automated Model Comparison

Thu, 01 May 2025 00:00:00 GMT

AutoDiff: A Scalable Framework for Automated Model Comparison Woo, Andrew Kyoungwan Post-training adaptations such as supervised fine-tuning, quantization, and reinforcement learning can cause large language models (LLMs) with identical architectures to exhibit divergent behaviors. However, the mechanisms driving these behavioral shifts remain largely opaque, limiting the reliability and interpretability of adapted models. AutoDiff is a scalable, automated framework for tracing model divergence on a per-neuron basis. It exhaustively profiles every feed-forward (MLP) unit across a pair of models, identifies the neurons with the largest activation gaps, and links these differences to downstream behavioral changes. The pipeline identifies exemplars that maximize between-model activation divergence and clusters the highest-gap neurons into an interpretable, queryable difference report. Proof-ofconcept experiments on GPT-2 small validate AutoDiff’s ability to rediscover synthetic perturbations without manual supervision. A larger case study on Llama3.1–8B contrasts the base model with several adapted variants, surfacing neurons whose behavioral shifts align with observed topic-level gains and losses. By uncovering these mechanistic divergences, AutoDiff transforms black-box model updates into actionable insights, enabling safer deployment, principled debugging, and interpretable model evaluation.

Schrödinger’s Carbon: Until Measured, Operational Emissions Remain Uncertain

Thu, 01 May 2025 00:00:00 GMT

Schrödinger’s Carbon: Until Measured, Operational Emissions Remain Uncertain Xia, Julia Rapidly improving generative artificial intelligence has led to significant investments in datacenter infrastructure, driving power demand, and raising environmental concerns. This has led to a growing body of research towards modeling embodied and operational carbon of datacenter servers across a variety of paradigms. However, most existing models take in deterministic inputs and output a singular average value that does not capture the inherent variability in estimating embodied and operational carbon emissions. Further, these average outputs obscure the impact of interacting factors, such as those related to deployment or software characteristics; each of which has its own underlying uncertainty distribution. This means in most cases, these averages do not accurately represent a particular server’s context. This thesis explicitly parameterizes and quantifies the full probabilistic distribution of operational carbon in AI inference tasks. It explores several factors of variability— deployment, spatiotemporal, and computational profile— and quantifies their impact on the overall carbon footprint through statistical and sensitivity analysis. While this work focuses on operational carbon, uncertainty propagation and understanding of variability should be used across a datacenter server’s entire life cycle. When this methodology is used alongside the existing uncertainty-aware embodied carbon measurements, it enables a holistic assessment from cradle to grave. This facilitates informed decision-making in server replacement, workload scheduling, hardware procurement, capacity planning, and more scenarios.

Ideator Explorer: Enhancing AI-Assisted Ideation through Interactive Visualization

Thu, 01 May 2025 00:00:00 GMT

Ideator Explorer: Enhancing AI-Assisted Ideation through Interactive Visualization Wen, Haoran Current AI-assisted ideation systems, often based on linear chat interfaces, struggle to help users effectively manage the complexity of creative exploration, hindering both divergent thinking across multiple paths and the convergent synthesis of ideas. This thesis introduces and evaluates Ideator Explorer, a human-AI ideation system built upon an interactive graph visualization interface designed to overcome these limitations. The core of the system is its spatial, tree-like representation of branching idea sequences. Formative user studies indicate that this visualization approach is preferred over chat interfaces for its organizational benefits and its effectiveness in helping users track parallel lines of thought during exploration. The spatial layout inherently supports both the exploration of diverse idea branches (divergence) and the identification of potential connections (convergence). This research focuses on the design and evaluation of this interactive graph interface, examining how its specific visualization and interaction techniques impact the user’s ability to navigate, organize, and develop ideas within complex ideation processes. The primary contribution is a novel, visually driven interface paradigm for human-AI collaboration that enhances the management and exploration of the creative solution space.

Type Checker for Annotated Assembly Programs

Thu, 01 May 2025 00:00:00 GMT

Type Checker for Annotated Assembly Programs Zanders, Julian The rise of speculative-execution attacks, such as Spectre, has presented a security challenge to developers. Speculation on secret data can expose it, but running without speculation is suboptimal for runtime. To fix this, researchers have been evaluating “smart” speculation schemes, which determine when to speculate and when not to in order to balance runtime with security. Our lab proposes Octal, a solution that utilizes software and hardware in tandem. Data values are marked as secret or public using type inference, and the veracity of inference is checked using a type checker. Then, hardware can separate the secret and public values. My contributions were to the type checker, as well as some scripting to evaluate results.

Real-Time Non-Line-of-Sight Imaging Using Single-Photon LiDAR

Thu, 01 May 2025 00:00:00 GMT

Real-Time Non-Line-of-Sight Imaging Using Single-Photon LiDAR Tsao, Nicholas Robust real-time imaging systems have allowed for many advances in robotics and autonomous navigation, though limited visibility in many real-world settings remains a significant challenge. Non-Line-of-Sight (NLOS) sensing allows for imaging systems to “see around corners", expanding their range of perception, providing access information for realtime decision-making. A promising approach to NLOS sensing is through single-photon LiDAR, which is commonly used for range-finding in many imaging systems. In addition to range-finding, single-photon LiDAR systems can provide a deeply rich data source in the form of photon count histograms after reflecting off scene geometry, capturing detailed information from multiple bounces. NLOS imaging can be achieved by parsing third-bounce light from such single-photon LiDAR sensors, which can be used for a variety of detection and localization tasks, and recent work has demonstrated capabilities in a wide range of applications. This work aims to further develop the NLOS imaging system by demonstrating a fully functional NLOS system using low-cost, consumer-grade SPAD hardware for real-time NLOS imaging, detection, and localization. We lay the ground work for NLOS imaging systems by developing infrastructure for NLOS processing in real-time, and we examine the potential for NLOS systems to operate on cheap hardware using data-driven approaches. Our work implements and demonstrates full end-to-end capacity for these NLOS imaging systems in a number of applications including person detection and localization, facilitating future research in this field and paving the way for NLOS integration into consumer devices.

Automated Finetuning via Sparse Autoencoders

Thu, 01 May 2025 00:00:00 GMT

Automated Finetuning via Sparse Autoencoders Sivakumar, Ragulan Currently, the field of interpretability is traditionally confined to diagnostics. However, this thesis presents a novel method using interpretability in sparse autoencoders to achieve better performance in small models via instruction finetuning. Specifically, we present UnderstandTune, an autonomous method for assembling high-quality instruction finetuning datasets with minimal human intervention, requiring only concise task descriptions rather than evaluation dataset distributions. Our empirical evaluations show that UnderstandTune consistently outperforms uninformed finetuning baselines across multiple benchmarks. Complementing this, Lalon introduces a mixture-of-informed-experts (MoIE) architecture that routes queries to specialized models independently finetuned via UnderstandTune. This modular approach achieves competitive performance against larger monolithic models in specialized domains, while utilizing fewer parameters, training examples, and computational resources. The framework’s modularity enables independent optimization of components from sparse autoencoders to MoIE routing mechanisms. This research demonstrates how interpretability can be used to enhance performance through intelligent data curation and suggests a new paradigm where interpretability and efficiency reinforce each other toward more capable, resource-efficient AI systems.

Generative Machine Learning Models for RNA Structure Prediction and Design

Thu, 01 May 2025 00:00:00 GMT

Generative Machine Learning Models for RNA Structure Prediction and Design Rubin, Dana Ribonucleic acid (RNA) is a fundamental molecule in biology, central to the regulation and execution of life’s most essential processes. Its diverse roles range from encoding genetic information to catalyzing biochemical reactions. Beyond its modern biological functions, RNA is also believed to have played a pivotal role in the origins of life which underscores the evolutionary significance of RNA. Unlocking the full potential of RNA research and design requires a deep understanding of the intricate relationship between RNA’s three-dimensional structure and sequence. Predicting RNA 3D structures remains a challenging problem due to the complexity of its folding landscape and the limited availability of high-resolution structural data. Inspired by recent advances in deep learning for protein folding and design, this thesis explores novel geometric and generative architectures for modeling RNA. We first present a systematic study on RNA structure prediction using equivariant neural networks within diffusion probabilistic models (DDPMs). Our folding model, named Klotho, captures local atomic interactions and structural features using SO(3)-equivariant message passing layers with a point cloud data representation. Ablation studies confirm that Klotho’s model performance scales with higher dimensionality and improves with enriching the input with secondary structure information and sequence embeddings from RNA foundation models. Building on this foundation, we introduce RiboGen, a multi modal deep learning model to jointly generate both RNA sequence and all-atom 3D structure. RiboGen integrates Flow Matching and Discrete Flow Matching within a unified multi modal representation and employs Euclidean Equivariant Neural Networks to learn geometric features. Our results demonstrate that RiboGen can generate chemically plausible, self-consistent RNA molecules, highlighting the potential of co-generative models to explore the sequence–structure landscape of RNA in a unified, data-driven framework. Together, these contributions advance the field of RNA modeling by offering scalable, symmetry-aware architectures for prediction and design. They lay the groundwork for future generative systems in RNA biology, therapeutic development, and biotechnological innovations.

Resilient Object Perception for Robotics

Thu, 01 May 2025 00:00:00 GMT

Resilient Object Perception for Robotics Shi, Jingnan A broad array of applications, ranging from search and rescue to self-driving vehicles, requires robots to perceive and understand the geometry of objects in the environment. Object perception needs to reliably work in a variety of scenarios and preserve a desired level of performance in the face of outliers and shifts from the training domain. Obtaining such a level of performance requires robust estimation algorithms that are able to identify and reject outliers, as well as techniques to continually improve performance of learningbased perception modules during test-time. In this thesis, we address these challenges by proposing (1) certifiably optimal solvers and a graph-theoretic framework that together help achieve state-of-the-art pose estimation performance even under high outlier rates, (2) self-supervised object pose estimators that can improve performance during test-time with accuracy comparable to state-of-the-art supervised methods, and (3) a test-time adaptation method for both object shape reconstruction and pose estimation without the need for CAD models. Throughout the thesis, we demonstrate that by using a variety of tools from optimization and learning, we can develop resilient object perception systems that perform reliably in a wide range of conditions.

Enhancing a Data-Centric Framework for Predictive Maintenance of Wind Turbines

Thu, 01 May 2025 00:00:00 GMT

Enhancing a Data-Centric Framework for Predictive Maintenance of Wind Turbines Pan, Raymond Predictive maintenance of wind turbines is a machine learning task aimed at minimizing repair costs and improving efficiency in the wind turbine and renewable energy industry. Existing machine learning solutions often fail to meet real-world deployment requirements due to fragmented pipelines, lack of domain integration, and reliance on black-box models. Zephyr, a data-centric machine learning framework, addresses these challenges by enabling Subject Matter Experts (SMEs) to incorporate their domain knowledge into the prediction process, and to leverage automated tools for labeling, feature engineering, and prediction tasks without requiring extensive technical knowledge. However, the current version of Zephyr still has limitations, including usability gaps and a reliance on external tools for certain steps. Case studies with real-world data from the renewable energy company Iberdrola demonstrate Zephyr’s potential to integrate domain expertise into wind turbine predictive maintenance (thus streamlining the process) but also expose a sub-optimal user experience. This thesis explores gaps in the current state of the Zephyr framework and proposes refinements to enhance its usability. Key improvements include the consolidation of current tooling and relevant external libraries into a single API, state management with careful logging and exception handling, and improved support for model evaluation. These enhancements aim to support seamless end-to-end predictive modeling workflows, and to provide a more refined and flexible user experience for the Zephyr user base.

Minimalist Approach to End-to-End Vision Language Navigation with Multi-Modal Foundation Model Features

Thu, 01 May 2025 00:00:00 GMT

Minimalist Approach to End-to-End Vision Language Navigation with Multi-Modal Foundation Model Features Mishra, Kartikesh Recent vision-language navigation (VLN) approaches leverage large models, prompt engineering, and/or explicit reasoning for instruction interpretation and agent guidance. We introduce MiniNav, a minimalist framework employing frozen vision-language foundation models as patch-wise feature extractors, avoiding data and compute heavy fine-tuning and cumbersome language model reasoning. Our lightweight control policies (∼ 10⁵ trainable parameters) are trained on a compact dataset of language-based specified navigational behaviors (∼ 10² runs, ∼ 10⁴ frames per behavior). We demonstrate generalization to novel objects and scenes, including direct real-world transfer, despite training on only two objects in a single simulated environment. Through its simple and scalable design, MiniNav provides an alternative to computationally intensive pipelines for robust real-world instruction-following. Our solution can provide a reference for evaluating the effective edge of more complex and larger VLN policies.

Strategic Sampling: A Framework for Enhancing Speed and Performance of Financial Fraud Detection Models

Thu, 01 May 2025 00:00:00 GMT

Strategic Sampling: A Framework for Enhancing Speed and Performance of Financial Fraud Detection Models Mitchell, Samuel Financial fraud detection is a high-stakes field where rapid inference is essential. While state-of-the-art fraud detection models vary in terms of architectural decisions and appear to exhibit unique computational bottlenecks, we highlight that their run-times are all dominated by extensive information-gathering steps. These steps involve aggregating information from a large set of nodes or edges within a graph, and these intensive steps are performed O(|V |) or O(|E|) times during an inference forward pass, on a graph with |V | nodes and |E| edges. We introduce Strategic Sampling, a general method to accelerate these information-gathering steps. Our approach tailors sampling strategies based on the specific objective function used in each model’s information-gathering process, selecting the most relevant pieces of information to use in each step. This ensures that critical information is retained while significantly reducing the amount of data processed, thus speeding up the computation. We conceptually demonstrate how Strategic Sampling can be applied to message-passing Graph Neural Networks, Graph Transformers, and TGEditor (a state-of-the-art graph editing algorithm). To showcase the effectiveness of our proposed Strategic Sampling method, we implement it in the TGEditor codebase. Our results show that Strategic Sampling not only significantly reduces computation time by more than an order of magnitude, but also improves the F1 score, enhancing both efficiency and performance. This study underscores the potential of Strategic Sampling to universally boost the performance of various financial fraud detection models, paving the way for faster and more accurate fraud detection.

Grain Boundary Solute Segregation in Vanadium

Sat, 01 Feb 2025 00:00:00 GMT

Grain Boundary Solute Segregation in Vanadium Ng, Daniel S. Vanadium alloys are a candidate structural material in nuclear fusion applications, where the presence of grain boundaries can improve mechanical properties and act as a sink for radiation- induced defects. Solutes with a thermodynamic preference to segregate to grain boundaries can stabilize them, making this a prime consideration for alloy design, but there are limited quantitative solute segregation data for vanadium. Based on results from an ab-initio computational framework for predicting the spectrum of grain boundary segregation energies across the periodic table, select nanocrystalline vanadium-based binary alloy systems were synthesized via mechanical alloying for targeted experiments characterizing differences in segregation strength. Scanning transmission electron microscopy and energy-dispersive x-ray spectroscopy measurements of solute concentrations in the grain boundary and bulk validate computational predictions of the average segregation strengths for different solutes, while showing inhomogeneous solute distributions along the grain boundary network that confirm the necessity for a spectral model that captures the behavior of site-specific segregation energies. After establishing the segregation behavior of different solutes in vanadium, the effects of solute segregation on other properties are examined. Heating experiments demonstrate that vanadium alloys containing strongly segregating species retain smaller grain sizes upon thermal annealing, indicating better grain boundary stability. The powder metallurgical route used produce these vanadium alloys requires a subsequent sintering step to densify powders into bulk parts for engineering applications, and dilatometry experiments reveal that that the addition of strongly segregating solutes also dramatically suppresses the sintering behavior. A kinetic analysis of the dilatometry data suggests that rapid grain boundary diffusion pathways that are necessary for effective sintering are obstructed by solute segregant, which has important repercussions for the processability of these alloys. Finally, microstructural characterization and nanohardness testing after ion-irradiation experiments demonstrate that the alloys with solute-stabilized grain boundaries are more resistant to nanovoid formation and radiation hardening. The work in this thesis advances our understanding of solute segregation and its effects in vanadium alloys, and highlights an approach for controlling grain boundaries that may facilitate future alloy design efforts for improved microstructural stability and radiation damage tolerance.

Modular Construction of Complex-Architected Bottlebrush Block Copolymers and Their Self-Assembly Behaviors

Sat, 01 Feb 2025 00:00:00 GMT

Modular Construction of Complex-Architected Bottlebrush Block Copolymers and Their Self-Assembly Behaviors Sun, Zehao Microphase-separated block copolymers are attractive materials for self-assembled nanolithography, yet there is a disconnect between the simple patterns commonly formed by block copolymers and the complex patterns required for many nanoscale applications, particularly in microelectronics. To meet this challenge, researchers have sought to design and build copolymer systems at ever-increasing levels of complexity in the (macro)molecular level, which promises to show emergent intriguing properties that are otherwise absent. However, the synthetic challenge as well as the vastly increased parameter space have obscured the systematic study of such complex systems. An efficient, modular synthetic route is thus highly desired for Lego-like molecular construction of property-decoupled, individually-tunable target materials. In this thesis, we will highlight the research endeavor in developing a multiblock Janus bottlebrush copolymer architecture as a novel platform for generation of diverse nanostructures that have been challenging to fabricate. The architecture, which features two orthogonal Janus domains, can be facilely constructed from corresponding building blocks by graft-through synthesis and can yield hierarchically engineerable phase-in-phase patterns. Surprisingly, the two constituent domains, though relatively independent of each other, behave significantly differently when combined together under certain circumstances. Their collective behavior gives rise to two low-symmetry mesh-like network phases (monoclinic and tetragonal respectively) that have not been observed in other soft materials before, which are of both fundamental and technological interest. Through a suite of experimental and computational study, we show that this peculiar phenomenon is an outcome of intrinsic molecular confinement, an emergent effect unique to multi-body, multi-hierarchy complex architectures. This work demonstrates that intrinsic molecular confinement is a viable path to bottom-up assembly of new geometrical phases of soft matter, extending the capabilities of block copolymer nanofabrication. As another example of modular synthesis, we will show an iterative polymerization methodology for controlled synthesis of bottlebrush copolymers with expanded compositional and architectural scope. When synergizing with other components, this strategy allows rapid access to functional materials that display different phase behavior when compared to the self-assembly of conventional copolymers. Our work introduced here is expected to facilitate the synthesis of complex functional copolymers, spark interest in the exploration of their property-function relationship, and enable more opportunities for their application in nanopatterning and other advanced materials.

Optimization Techniques for Trustworthy 3D Object Understanding

Thu, 01 May 2025 00:00:00 GMT

Optimization Techniques for Trustworthy 3D Object Understanding Shaikewitz, Lorenzo Franceschini Autonomous machines require reliable 3D object understanding to interpret and interact with their environment. In this thesis, we consider two tightly coupled 3D object understanding problems. Shape estimation seeks a consistent 3D model of an object given sensor data and some set of priors. Pose estimation seeks an estimate of the object’s position and orientation relative to an invariant shape frame. In general, these problems are non-convex and thus difficult to solve. We present algorithms which nonetheless solve shape and pose estimation efficiently and with assurances in terms of of optimality, uncertainty, or latency. We begin in the multi-frame tracking setting, where we propose the certifiably optimal estimator CAST⋆ for simultaneous shape estimation and object tracking. CAST⋆ uses 3D keypoint measurements extracted from an RGB-D image sequence and phrases the estimation as fixed-lag smoothing. Temporal constraints enforce rigidity and continuous motion. Despite the non-convexity of this problem, we solve it to certifiable optimality using a smallsize semidefinite relaxation. We also present a compatibility-based outlier rejection scheme to handle outliers, and evaluate the proposed approach on synthetic and real data. Next, we focus on estimating the pose of an object given its shape and a single RGB image (no depth). Assuming only bounded noise on 2D keypoint measurements (e.g., from conformal prediction), we derive an estimator for the most likely object pose which uses a semidefinite relaxation to initialize a local solver. We pair this with an efficient uncertainty estimation routine which relies on a generalization of the S-Lemma to propagate keypoint uncertainty to high-probability translation and rotation bounds. The high-probability bounds hold regardless of the accuracy of the pose estimate, and are reasonably tight when tested on the LineMOD-Occluded dataset. Lastly, we propose a sub-millisecond solution to simultaneous estimation of object shape and pose from a single RGB-D image. Our approach converts the first-order optimality conditions of the non-convex optimization problem to a nonlinear eigenproblem in the quaternion representation of orientation. We use self-consistent field iteration to efficiently arrive at a local stationary point, finding solutions more than an order of magnitude faster than Gauss-Newton or on-manifold local solvers on synthetically generated data.

Joint Localization and Synchronization via User Cooperation in Non-Terrestrial Networks

Thu, 01 May 2025 00:00:00 GMT

Joint Localization and Synchronization via User Cooperation in Non-Terrestrial Networks Morrison, James C. Next-generation (xG) wireless networks require accurate localization and synchronization for efficient resource management and emerging applications. Non-terrestrial networks (NTN) with low Earth orbit (LEO) satellites offer a promising alternative for positioning, navigation, and timing (PNT) by providing diversity and increasing the signal-to-noise ratio (SNR) over global navigation satellite systems (GNSS). However, the primary challenge in NTNbased localization with LEO satellites is the lack of precise clock synchronization, which introduces biases in time-of-arrival (TOA) measurements and limits localization accuracy. This paper introduces a joint cooperative localization and synchronization (JCLS) framework that addresses this challenge through spatiotemporal cooperation, soft information, and simultaneous synchronization. Furthermore, we propose a three-step algorithm for performing JCLS. The first step calculates a coarse position estimate using TOA measurements and the Gauss-Newton method. Then, this coarse estimate is updated using the LevenbergMarquardt method which performs joint localization and synchronization. Finally, we derive a soft information-based filter that is used to continuously refine the position and clock error estimates as new measurements are available. We characterize the fundamental performance limits of JCLS using Fisher information, which offers insight into its localization and synchronization accuracy bounds. Furthermore, simulation results based on TOA measurements of the 3rd Generation Partnership Project (3GPP) 5G New Radio positioning reference signal (PRS) demonstrate that the proposed algorithm for JCLS significantly improves localization and synchronization accuracy compared to non-cooperative methods.

Multipartite Quantum Clock Synchronization Via Collective Symmetric States

Thu, 01 May 2025 00:00:00 GMT

Multipartite Quantum Clock Synchronization Via Collective Symmetric States Keskin, Ufuk This thesis investigates multipartite quantum clock synchronization (QCS) tasks using a class of quantum states, called collective symmetric (CS) states, which generalize Dicke and N00N states. Employment of CS states in previous QCS procedures is shown to improve synchronization performance in various network scenarios. The focus of the paper is on QCS procedures that, after the distribution of quantum states, rely exclusively on local operations and classical communication (LOCC), ensuring compatibility with highly noisy quantum channels. Two synchronization scenarios are considered: (i) synchronization between the two nodes of an arbitrarily chosen pair of nodes, and (ii) global synchronization where all nodes wish to synchronize their clocks to a common average time. First, a framework in which the previous procedures operate employing the CS states is introduced. Using such framework, possible limitations of the QCS procedures in terms of estimation ambiguity and lack of robustness are pointed out. Second, a procedure referred to as the tactical delay procedure (TDP) is proposed for each of the two synchronization scenarios. The TDP resolves the mentioned limitations and outperforms the state-of-the-art multi-partite QCS procedures in terms of synchronization precision without requiring additional quantum resources.

Accelerating Embedded HOWFSC Algorithms

Thu, 01 May 2025 00:00:00 GMT

Accelerating Embedded HOWFSC Algorithms Eickert, Brandon The quest to directly image planets of other solar systems demands not only state-of-the- art coronagraphs, but also places extreme performance demands on space-based processors. Direct imaging requires precise wavefront control to acquire the 1010 contrast necessary to reveal a dim, Earth-like exoplanet. This precise level of control is only possible if high-order wavefront sensing and control (HOWFSC) algorithms are executed with enough speed to offset wavefront error accumulation. Of the many aspects that make high-contrast imaging difficult, a central bottleneck is the speed at which we can run these algorithms. At the center of this work, we aim to accelerate the execution of two foundational HOWFSC algorithms: optical modeling and Electric Field Conjugation (EFC). Optical modeling underpins both Jacobian-based EFC, and a relatively new variant of EFC, called adjoint-based EFC. The two main contributions of this thesis are to port bottleneck HOWFSC algorithms to the relevant computing environments, and quantify speedups attained by both algorithm choice and implementation optimization. This work explores the acceleration of optical modeling for a vector vortex coronagraph through the use of the FFTW library, and the acceleration of EFC by implementing adjoint-based EFC in an embedded context. We utilize functional analogs to radiation-hardened processors, using the NXP T1040 in place of the BAE RAD5545, and the NXP LS1046 in place of the LS1046-Space. We find that the FFTW library enabled a factor of six speedup for 4096 × 4096 fast Fourier transforms (FFTs), and a factor of five for 2048 × 2048 FFTs. With these significant speedups, the bottleneck within the vortex operations of the optical model shifts from the FFT to matrix multiplication. We additionally time the execution of the underlying routines of Jacobian-based EFC and AD-EFC to estimate that AD-EFC is 46 times faster than Jacobian-based EFC. Despite these speedups, AD-EFC is still a factor of 124 away from 100-second latency for our specific optical model. These results demonstrate that one to two orders of magnitude of speedup must be attained by either further optimizing algorithm implementations, or exploring other parallelization strategies, computing architectures, and mission paradigms to achieve a latency on the order of 100 seconds.

Formalizing Causal Models Through the Semantics of Conditional Independence

Thu, 01 May 2025 00:00:00 GMT

Formalizing Causal Models Through the Semantics of Conditional Independence Zhang, Anna Many foundational tools in causal inference are based on graphical structure and can involve complex conditions that obscure the underlying causal logic. Given the inherent complexity and subtlety of cause-and-effect phenomena, establishing formal guarantees about these tools is both challenging and important. This thesis presents a semantics-driven formalization of causal models within the Coq proof assistant, enabling precise, mechanized reasoning about causal relationships. Central to this work is a new function-based definition of conditional independence, which captures how changes propagate through a causal graph. We prove that this semantic notion is equivalent to the standard graphical criterion of d-separation, thereby establishing a rigorous bridge between structural and semantic interpretations of independence. The formalization includes a library of graph-theoretic and causal-reasoning tools, encompassing key concepts such as mediators, confounders, and colliders. By linking the syntactic and semantic perspectives on causality, this work lays a robust foundation for formally verifying causal assumptions and guiding experimental design.

Contextual Knowledge Sharing in Multi-Agent Long-Horizon Planning Settings with Centralized Communication and Coordination

Thu, 01 May 2025 00:00:00 GMT

Contextual Knowledge Sharing in Multi-Agent Long-Horizon Planning Settings with Centralized Communication and Coordination Zhang, Jackson Embodied multi-agent systems, comprising autonomous agents interacting within shared environments, enable intelligent, collaborative solutions for tasks requiring real-time coordination and adaptability. While applications span diverse fields, from disaster response to healthcare, planning in these systems remains challenging due to partial egocentric observations and limited environmental awareness. This work addresses these challenges by introducing a software module that synthesizes a shared world state from individual agent views, maintaining spatial information about objects and agents to support more effective joint action planning. Integrated into the LLAMAR framework, this module aims to improve planning accuracy and efficiency. The proposed approach is evaluated using metrics such as success rate, transport efficiency, and coverage performance. Our evaluation demonstrates that utilizing a perfect (oracle-generated) world state significantly enhances planning effectiveness. Notably, under these ideal conditions, the success rate of the LLAMAR planner improved by over 16%. These findings underscore the critical impact of accurate world state representation on multi-agent performance and highlight the potential for significant advancements in collaborative task execution in dynamic, unstructured settings.

Characterization, Processing, and Synthesis of Extreme-Performance Continuous Carbon Nanotube Network Composites

Sat, 01 Feb 2025 00:00:00 GMT

Characterization, Processing, and Synthesis of Extreme-Performance Continuous Carbon Nanotube Network Composites Durso, Michael Nathan Continuous carbon nanotube (CNT) networks are an emerging, hierarchically-structured, and commercially available nanomaterial built from countless CNT nanocrystals. These macroscopic yarn materials promise to bridge the gap between microscopic CNT fibers – which are well-known for their superlative material properties – and human-scale fiber reinforcements for extreme-performance composites. Yet because the constituent CNTs interact only via intermolecular forces, network properties fall short of their building blocks. Although these materials show promise as reinforcement in composites, the networks’ low-permeability and tortuous nanoporous structure renders imbibition with liquids like a polymer matrix or surface functionalizing agents challenging. Thus, traditional composite fabrication strategies can be ineffective when applied to CNT yarns, especially commercial products subject to proprietary microstructural manipulation. Using commercially-available CNT yarns fabricated through floating-catalyst chemical vapor deposition (FCCVD) as model systems, we first explore yarn characteristics which are unique to their hierarchical, bundled-fiber structure, placing focus on the oxygen-rich amorphous carbon phase found in pre-densified, chemically-stretched yarns. A green hydrothermal technique is explored to remove this phase from the surface level inward, allowing for purification and improved infiltrability. However, we find this phase is distinct from previously-reported amorphous carbons found in CNTs, showing it behaves as a matrix which may improve polymer bonding. An analysis of imbibition and fluid transport in these CNT yarns finds that while infiltration of low-viscosity liquids like water is thermodynamically-favored, it is limited when surpassing the threshold of capillary pore percolation. Nevertheless, infiltration in lower-density networks is not only observed, but exploited through the demonstration of dielectric heating in a microwave reactor, where we show fluid imbibed within the network can be boiled to induce swelling and exfoliation of CNT bundles (or conversely, this may be avoided) through optimization of the heating parameters and solvent. Next, with a firm understanding of the yarn networks’ properties and the impact of various processing effects, we demonstrate two techniques of producing polymer in-situ using dissolved monomers to side-step slow infiltration. The first technique is in-situ interfacial polymerization (ISIP), which is adapted to the yarns studied in this work to yield polyetherimide–CNT yarn composites. When applied to chemically-stretched yarn, specific strengths as high as 2.2 GPa/(g-cm3) are achieved in the flexible and durable yarn composite. We show parameters and conditions which maximize tensile properties and challenges associated with the rapid nature of the process, concluding with the successful demonstration of a roll-to-roll fabrication scheme for producing arbitrary amounts of polymer. In our second technique, we produce extreme-performance polyimide and polybenzimidazole composites through green in-situ polymerizations (ISSP) in CNTs and macroscopic fiber networks. This approach utilizes superheated water and alcohol as a powerful medium to disperse monomers and initiate polymerization of high-performance coatings within a porous network. We demonstrate ISSP-CNT composites with variable coating morphologies (conformal, shish-kebab, etc.), in-air stability to over 500°C, and doubled specific stiffness and specific strength. Finally, we validate the multifunctional behavior of polyimide-CNT composites by showing a strong, flexible composite can store energy and behave as a free-standing battery electrode.

Parametric Study of Novel Passive Thermal Control Technology for Spacecraft

Thu, 01 May 2025 00:00:00 GMT

Parametric Study of Novel Passive Thermal Control Technology for Spacecraft Shafer, Emma Thermochromic variable emissivity materials (VEMs) are a relatively new passive thermal control technology used for spacecraft radiators. VEMs passively change their emissivity based on their temperature, with VEMs having low emissivity at low temperatures and high emissivity at high temperatures. This property of VEMs allows for spacecraft to have reduced heater power and less extreme temperature swings without adding active thermal control systems. There is a potential for VEM technology to become more widely used in spacecraft radiators. Because thermochromic VEMs are still a relatively new technology, there has not yet been a study with a parametric sweep of some possible VEM profiles and common spacecraft parameters to determine the best-case uses of particular VEM profiles. This thesis models a single-node spacecraft in an equatorial low Earth orbit, varying the spacecraft’s shape, surface area, and thermal mass using Thermal Desktop. The temperature history of the spacecraft in orbit, particularly its orbit minimum temperature, orbit maximum temperature, orbit average temperature, and orbit temperature range, is recorded, and twelve VEM profiles are compared against default black and white paint materials to see how the twelve VEM profiles change orbit minimum temperature, maximum temperature, average temperature, and temperature range. The desired outcome is for the VEMs to reduce the temperature range the most compared to black or white paint while keeping temperatures within typical temperature requirements for spacecraft components. It is found that, compared to white paint, VEMs always increase the orbit minimum temperature, maximum temperature, average temperature, and temperature range across all nodal thermal masses and surface areas studied. For spacecraft with lower surface areas, having only white paint decreases the temperature too much for typical spacecraft components, so even though white paint always decreases temperature range compared to VEMs, it is recommended to have VEMs instead of white paint for lower surface area spacecraft due to VEMs being better than white paint at keeping components within typical temperature requirements. When VEMs are compared to black paint, it is found that black paint has lower minimum temperatures and greater maximum temperatures than all VEMs at greater surface areas. For lesser surface areas, the node covered in black typically has minimum and maximum temperatures in the middle of the VEMs’ minimum and maximum temperatures. For all surface areas and thermal masses, the average temperature of the black node is typically in the middle of the average temperatures of the nodes with VEMs; in relation to the VEMs’ average temperatures, the black average temperature decreases as node height increases. For all node heights and thermal masses, VEMs always decrease the temperature range compared to black. VEMs are shown to be better than black paint in having spacecraft components stay within typical temperature requirements, and which VEM to choose depends on what the specific spacecraft component is and its specific temperature requirements. The biggest difference in individual VEM profiles compared to each other is the orbit average temperature; the lower the VEM’s transition temperature, the lower the average temperature. Only at the greatest nodal surface areas and smallest nodal heights is there a significant difference in temperature range between individual VEM profiles; typically, the lower the transition temperature of the VEM, the less its temperature range. Future work includes expanding on the parameters studied and studying spacecraft in different orbits, different spacecraft shapes, and different VEM profiles.

Path Planning for Autonomous Sailing Vessels: Developing Robust and Efficient Survey Strategies

Thu, 01 May 2025 00:00:00 GMT

Path Planning for Autonomous Sailing Vessels: Developing Robust and Efficient Survey Strategies Ahlers, Matthew C. Autonomous sailing vessels offer a promising solution for maritime research, offering low maintenance and sustainable platforms for environmental monitoring and data collection. These vessels utilize wind power, eliminating the need for conventional fuel and enabling long-duration operations with minimal environmental impact. Their applications range from oceanographic studies to maritime surveillance, where persistent and autonomous data collection is essential. This thesis explores the challenges and methodologies associated with path planning for autonomous sailing, particularly in the context of survey operations. Unlike traditional motorized vessels, sailing autonomy must account for wind variability, sail dynamics, and limited maneuverability, requiring specialized path-planning techniques to ensure efficient and reliable navigation. The research investigates various sail and hull configurations, the dynamics of windpowered propulsion, and the application of autonomy frameworks such as MOOS-IvP. A key focus is on optimizing continuous coverage path planning (CPP) to maximize efficiency while adapting to environmental constraints. By integrating real-time wind data and vessel performance characteristics, the study refines survey strategies that enhance mission effectiveness. Different survey strategies are implemented and evaluated using both simulation and real-world testing on the Charles River. These trials demonstrate the feasibility of fixed-path decomposition approaches and adaptive moving horizon control methods, evaluating methods with the impact of wind conditions on autonomous sailing performance. The results contribute to the development of robust and efficient survey strategies that improve the autonomy and reliability of wind-powered marine vessels.

Census-Based Population Autonomy for Marine Robots: Theory and Experiments

Thu, 01 May 2025 00:00:00 GMT

Census-Based Population Autonomy for Marine Robots: Theory and Experiments Paine, Tyler Collaborating groups of robots show promise due in their ability to complete missions more efficiently and with improved robustness, attributes that are particularly useful for systems operating in marine environments. A key issue is how to model, analyze, and design these multi-robot systems to realize the full benefits of collaboration even with limited communication, a challenging task since the domain of multi-robot autonomy encompasses both collective and individual behaviors. This thesis presents a layered model of multi-robot autonomy that uses the principle of census, or a weighted count of the inputs from neighbors, for collective decision-making coupled with multi-objective behavior optimization for individual decision-making. The census component is expressed as a nonlinear opinion dynamics model and the multi-objective behavior optimization is accomplished using interval programming. This model can be reduced to recover foundational algorithms in distributed optimization and control, while the full model enables new types of collective behaviors that are useful in real-world scenarios. To illustrate these points, a new method for distributed optimization of subgroup allocation is introduced where robots use a gradient descent algorithm to minimize portions of the cost functions that are locally known, while being influenced by the opinion states from neighbors to account for the unobservable costs. With this method the group can collectively use the information contained in the Hessian matrix of the total global cost. In addition, the critical issue of controlling subgroup size to minimize a collective cost signal is addressed, an initial step toward establishing a general definition of controllability of the nonlinear opinion dynamics model. The utility of this model is experimentally validated in three categorically different experiments with fleets of autonomous surface vehicles: an adaptive sampling scenario, a high value unit protection scenario, and a competitive game of capture the flag.

Reliable and Generalizable Real-World Planning with LLM-based Formalized Programming

Thu, 01 May 2025 00:00:00 GMT

Reliable and Generalizable Real-World Planning with LLM-based Formalized Programming Hao, Yilun While large language models (LLMs) have recently demonstrated strong potential in solving planning problems, LLMs, as zero-shot planners themselves, are still not capable of directly generating valid plans for complex planning problems such as multi-constraint or long-horizon tasks. This motivates the needs to develop a robust and reliable planning system for complex real-world planning problems. Furthermore, many frameworks aiming to solve complex planning problems often rely on task-specific preparatory efforts, such as task-specific in-context examples and pre-defined critics or verifiers, which limits their cross-task generalization capability. This motivates the needs to extend the robust and reliable planning systems to have strong generalization capability. In this thesis, we first develop an LLM-based planning framework that formalizes and solves complex multi-constraint planning problems as constrained satisfiability problems and can reliably identify the unsatisfiable cores for unsatisfiable requirements, provide failure reasons, and offers personalized modification suggestions. Then, we generalize the paradigm by proposing a general-purpose framework that leverages LLMs to capture key information from planning problems and formally formulate and solve them as optimization problems from scratch, with no task-specific examples needed. Comprehensive experimental results have shown that our frameworks significantly outperform the baselines and have strong performance across tasks and LLMs.

Experimental Quantification of the Phonon Drag Deformation Mechanism in Metals at Extreme Strain Rates

Sat, 01 Feb 2025 00:00:00 GMT

Experimental Quantification of the Phonon Drag Deformation Mechanism in Metals at Extreme Strain Rates Dowding, Ian Extreme strain rate deformations, above 10⁶ s⁻¹, are seen across many fields of science and engineering; from meteorite impacts and impact induced crystallographic phase changes to high-speed machining and additive manufacturing. Despite the range of applications, many common high-rate impact experiments are intrinsically limited to strain rates of only 10⁴ s⁻¹ before complicating the material deformation with a superimposing state of shock due to high impact pressures. However, recent advances in optically driven microballistics using laser induced projectile impact tests have provided a new quantitative look into extreme mechanics of materials, at rates above 106 s-1 and well below the onset of shock effects. As deformation strain rates increase, additional strengthening mechanisms in metals become available, leading to a change in the underlying physics of dislocation motion and an increase in strength. This thesis first explores the mechanical properties of pure metals when deformed at extreme strain rates − both in ambient conditions and elevated temperatures. Using an array of complimentary characterization methods, two independent measurements of strength, the dynamic strength and dynamic hardness, are assessed. As the temperature is increased from ambient, the strength and hardness of pure metals both increase an appreciable amount. At these deformation rates, conventional thermal softening effects are now in competition with anti-thermal hardening that arises from ballistic transport of dislocations from phonon interactions in the crystal lattice. These effects are quantified systematically and it is shown that the anomalous thermal strengthening seen is, thermodynamically and kinetically, the expected form of plasticity under these impact conditions. Next, the limits of where this anomalous thermal strengthening occur in metals are investigated. First, solute elements are added to pure Ni to evaluate how additional dislocation pinning mechanisms effect the strength at ambient and elevated temperatures during extreme strain rate deformations. The strengthen increase due to solute pinning of dislocations is additive to the other strengthening mechanisms, yet thermally controlled, which provides a transition from ballistic transport of dislocations to thermally activated strengthening at a critical concentration of solutes. Finally, the upper bound of temperature for dislocation phonon drag strengthening is assessed. While it was shown that pure metals increase strength with increasing temperature, this “hotter-is-stronger” trend breaks down as the temperature approaches the melting point of the metal. Using Sn, due to its low melting temperature, the breakdown from “hotter-is-stronger” to “hotter-is-softer” as the initial substrate temperature approaches the melting temperature is systematically explored.

Concentration-Dependent Thermodynamics and Kinetics in Lithium-Metal Battery Electrolytes: Implications for Coulombic Efficiency

Sat, 01 Feb 2025 00:00:00 GMT

Concentration-Dependent Thermodynamics and Kinetics in Lithium-Metal Battery Electrolytes: Implications for Coulombic Efficiency Plaza Rivera, Christian O. Lithium (Li)-metal batteries (LMBs) present a promising avenue for high-energy applications. However, their practical adoption is constrained by challenges such as dendrite formation and unstable interphases. This study investigates the intricate interplay between electrolytedependent thermodynamics, kinetics, and transport properties in LMBs, focusing on the concentration effects in fluoroethylene carbonate (FEC) and 1,2-dimethoxyethane-based electrolytes containing lithium bis(fluorosulfonyl)imide. Due to FEC’s unique properties, these electrolytes facilitate significant upshifts in the Li redox potential and contribute to stable interphases and voltage profiles. Our findings reveal that the redox potential is primarily governed by the solvent’s electron-donating ability, reflecting underlying solvation dynamics, while the electrolyte permittivity influences reaction entropy trends. The results show entropy changes from increased molecular disorder at moderate concentrations to reduced entropy in highly concentrated regimes, driven by the formation of ion–solvent complexes. Kinetic analyses demonstrate a volcano-shaped dependence of exchange current density on concentration, centered at 2 M. Two prevailing perspectives propose that either kinetic–transport interplay or thermodynamic properties govern Coulombic efficiency (CE). However, separating these contributions is complex, since both higher exchange current density and upshifts in the Li redox potential enhance CE. Furthermore, CE strongly aligns with the combined effects of kinetics, thermodynamics, and transport, emphasizing the need for a holistic electrolyte design approach. Optimizing these three factors makes it possible to stabilize the interphase, promote uniform Li deposition, and elevate the overall safety and performance of next-generation LMBs.

1500W High Voltage DC-DC Converter for Electroaerodynamic Aircraft Applications

Thu, 01 May 2025 00:00:00 GMT

1500W High Voltage DC-DC Converter for Electroaerodynamic Aircraft Applications Shevgaonkar, Mihir Electroaerodynamic (EAD) propulsion is a novel form of propulsion that is nearly silent and has no moving parts. The first functional untethered heavier-than-air EAD aircraft had an endurance of 90 seconds and could only fly in a straight line. To enable a practical fixed wing EAD aircraft that can fly outdoors with a payload for an extended period of time, improved power conversion technology is necessary. Prior work specifies a practical EAD aicraft as one with an endurance of 10 minutes, a payload capacity of 200 g, and full controllability. This work explores methods of increasing the specific power of power converters for EAD aircraft from 1.15 kilowatts per kilogram to over 2.0 kilowatts per kilogram. Such an increase can be achieved by utilizing magnetics integration and thermal management techniques, as well as adjustments in the operating point of the power converter. The power converter for the first generation EAD aircraft had an input voltage of 200 V, an output voltage of 40 kV, an output power of 600 W, a specific power of 1.15 kilowatts per kilogram, and an efficiency of 85 percent. In this work, a power converter with an input voltage of 200 V, an output voltage of 20 kV, an output power of 1476 W, a specific power of 2.7 kilowatts per kilogram, and an efficiency of 96 percent was demonstrated to work for a 40 second duration. At the end of the test, device temperatures continued to increase, so it has not been proven that the converter can work in thermal steady state as required for a 10 minute flight. Future work would involve modifying the test setup to allow for adequate ventilation of the ambient air around the converter, as well as modifying the converter with adequate thermal management so as to enable operation under thermal steady state.

Feasibility Analysis and Fuel Burn Benefits of Relaxing Constraints in High Altitude Cruise

Thu, 01 May 2025 00:00:00 GMT

Feasibility Analysis and Fuel Burn Benefits of Relaxing Constraints in High Altitude Cruise Cezairli, Mina Operational interventions, such as enabling more fuel-efficient trajectories, are desirable in mitigating the environmental impact of air travel due to their relatively fast implementation potential. In particular, the vertical inefficiency arising from the altitude stratification in the airspace can be mitigated by relaxing vertical constraints. The feasibility of vertical flexibility is evaluated by quantifying the rate of close encounters and the frequency of alerts that would be needed to prevent them. Substantial diurnal variability in the number of close encounters was found in the airspace, with lower rates of events during the nighttime period. Furthermore, regional differences among Air Route Traffic Control Centers were observed in the number of close encounters. The frequency of controller intervention events that would have to occur was evaluated at 25 NM and 50 NM alerting distance levels, and it was found that, given sufficient technological capabilities for alerting at the 25 NM reaction distance, most centers would have fewer than 10 alerts per hour during the nighttime period. Boston, Miami, and Seattle appeared especially promising, with approximately one alert per hour for each region. Finally, the potential fuel benefit from enabling vertically optimal trajectories was estimated to be up to 100,000 gallons of fuel savings per month in the case of a CONUS-wide nighttime implementation.

Risk Management in Air Traffic Applications: Data-Driven Modeling, Prediction, and Generation of Realistic Weather Disruptions and Other Unfavorable Conditions

Thu, 01 May 2025 00:00:00 GMT

Risk Management in Air Traffic Applications: Data-Driven Modeling, Prediction, and Generation of Realistic Weather Disruptions and Other Unfavorable Conditions Zhang, Joseph Understanding the interaction between weather and disruptions in complex air transportation network is important to the design and evaluation of preemptive measures and responses taken by air traffic managers. However, the occurrence of disruptive weather events is often rather limited compared to the amount of data available for nominal operations. Additionally, in large-scale systems with many known and unknown confounding factors, it can be difficult to identify the relevance of existing data to different underlying distributions of interest. Furthermore, existing work generally follows a frequentist paradigm in predicting disruptions based on weather, and does not easily lend itself to inferring the causes of disruptions, which can be important both in building models and using them to make predictions, and generate test cases to stress-test proposed design decisions. In this thesis, we develop a hierarchical Bayesian model for air traffic network operations, and investigate methods for learning these models in data-constrained settings, by extend existing work on retrospectively analyzing failures. We also include a guiding case study performed on LaGuardia Airport, in which a generative model is developed for the interaction between weather conditions and airport-level parameters within a single airport, trained on unlabeled historical data, and evaluated by simulating disruptions on historical schedules.

Interposing the Syscall Boundary: Transparent Python Execution in SigmaOS

Thu, 01 May 2025 00:00:00 GMT

Interposing the Syscall Boundary: Transparent Python Execution in SigmaOS Wu, Ivy σOS aims to provide both serverless and stateful support to cloud applications while maintaining strong isolation, security, and efficient startup times and scheduling among multiple users. While σOS and its container startup times have been successfully benchmarked for tasks written, compiled, and statically linked in Golang and Rust, it currently lacks support for other languages, including interpreted ones like Python. To bridge this gap, this paper presents the first integration of an interpreted language into σOS, enabling native Python support without compromising the system’s core principles. Our design, σPy, achieves this through three key ideas: (1) system call interposition via LD_PRELOAD to enable just-in-time dependency management, where Python libraries are fetched on-demand from tenant-specified AWS S3 buckets, avoiding overhead during container initialization; (2) a multi-layered mount namespace that spans the local machine, a per-realm Docker container, and a per-proc σcontainer, enabling efficient dependency caching at the per-tenant granularity; and (3) a hybrid C++, C, and Python API layer that bridges σOS’s Protobuf-based RPC system with Python’s dynamic types. Preliminary benchmarks demonstrate that σPy achieves performance comparable to that of compiled languages like Golang when interacting with the σOS API, with only 0.2 - 0.3 additional milliseconds of overhead on all tested API calls, validating the success of Python programs on the σOS architecture.

Simulating LLM Runtime Latency

Thu, 01 May 2025 00:00:00 GMT

Simulating LLM Runtime Latency Wang, Sarah Y. Large Language Models (LLMs) are expensive to run and can incur high latencies. Each LLM application has its own cost and latency targets. For example, AI voice assistants operate under low latency objectives, while large document batch processing jobs are typically cost-sensitive. However, navigating these trade-offs is not trivial, as LLM latency is highly task– specific and depends on factors such as the offered query load, the hardware configurations, request properties, and various model characteristics. To support the user in configuring their deployment according to their application needs, we introduce vLLMSim, an accurate simulator that estimates the latency of a given workload on different hardware configurations. vLLMSim advances two key avenues toward latency-aligned LLM deployments. First, the simulated latency metrics inform the user’s model and hardware choice, so they can use a configuration that is ideal for their workload. Second, our simulator enables researchers to quickly test latency-improving ideas, bypassing the need for time-consuming implementations before validating their effectiveness. In fact, vLLMSim is already used in two research projects with the goal of reducing latency and cost of LLM inference. In this thesis, we show how vLLMSim’s design allows it to accurately support the use cases above, while providing highly accurate runtime predictions. To support hardware exploration without GPU access, vLLMSim provides precomputed performance profiles that are sufficient to accurately simulate the user’s workload. The simulator code can be found here, and the instrumented vLLM code for creating profiles can be found here.

Methods for Latent Space Interpretation via In-the-loop Fine-Tuning

Thu, 01 May 2025 00:00:00 GMT

Methods for Latent Space Interpretation via In-the-loop Fine-Tuning Wen, Collin With language models increasing exponentially in scale, being able to interpret and justify model outputs is an area of increasing interest. Although enhancing the performance of these models in chat mediums has been the focus of interaction with AI, the visualization of model latent space offers a novel modality of interpreting information. Embedding models have traditionally served as a means of retrieving relevant information to a topic by converting text into a high-dimensional vector. The high-dimensional vector spaces created via embedding offer a way to encode information that captures similarities and differences in ideas, and visualizing these nuances in terms of meaningful dimensions can offer novel insights into the specific qualities that make two item similar. Leveraging fine-tuning mechanisms, dimension reduction algorithms and Sparse Autoencoders (SAEs), this work surveys state-of-the-art techniques to visualize the latent space in highly interpretable dimensions. ConceptAxes, derived from these techniques, is a framework is provided to produce axes that can capture high-level ideas that are ingrained into embedding models. ConceptAxes with highly interpretable dimensions allow for better justification for the latent space and clusters. This method of increasing embedding transparency proves valuable in various domains: (1) AI-enhanced creative exploration can be more guided and customized for a particular experience and (2) high-level insights can be made more intuitive with vast text datasets.

Commanding, Telemetry, and Software Strategy for CubeSat Laser Infrared CrosslinK (CLICK) Mission

Thu, 01 May 2025 00:00:00 GMT

Commanding, Telemetry, and Software Strategy for CubeSat Laser Infrared CrosslinK (CLICK) Mission Whitmore, Garrett This work outlines the software-related requirements necessary for successful operations of the NASA-sponsored Cubesat Laser Infrared CrosslinK (CLICK) B/C mission [1] [2]. This twin-cubesat mission will demonstrate peer-to-peer laser-communication capabilities novel at this small terminal scale. Optical laser communication terminals can have lower Size, Weight, and Power (SWaP) compared with traditional radio communication, as well as fewer licensing regulations and improved link security. CLICK-B/C follows from CLICK-A, a risk-reduction mission that successfully performed laser downlink with a ground station at MIT [3]. In addition to downlink, B/C will perform crosslink experiments at a data transmission rate over 20 Mbps at ranges between 20 and 580 km in Low-Earth Orbit (LEO). This thesis focuses on the software related to the function of the satellite payload, in particular, the improvements and additions made to the operating system, software systems that were ported over from CLICK-A, the integration and testing of these subsystems, and analyses done to prepare for in-flight operations before launch. An overview of the MIT & UF payload hardware and electronics is given before detailing interactions with components as necessary. A deep dive into the payload software libraries, internal and external communication channels, and operating system build details are given. A description of functional testing and its results are laid out as well as a template crosslink experiment script and further specifications for mission-related analyses and pre-launch preparations. This work on software upgrades, verification, and examination is necessary for CLICK-B/C to reach its stated mission goals, here on Earth and in its orbit.

Foundational Verification of Running-Time Bounds for Interactive Programs

Thu, 01 May 2025 00:00:00 GMT

Foundational Verification of Running-Time Bounds for Interactive Programs Tockman, Andrew The field of formal methods has a rich history of practical application in verification of the correctness of software. Existing verification tooling operates at a wide range of rigor, from proving relatively weak properties via traditional static analysis to powerful theorem provers that can express very precise specifications. It is sometimes desirable to prove properties about programs that make reference to not just semantic behavior but also to other metaproperties of the program’s execution, such as runtime or I/O histories. There is also a wide variety of existing tooling for proving bounds on program runtime. However, there is no prior work on a maximally rigorous verification system that can prove predicates involving all of semantic behavior, runtime, and I/O. Our contribution is exactly that – we extend the existing Bedrock2 framework, which implements a C-like systems language within a powerful proof engine together with a verified compiler capable of expressing arbitrary proof conditions involving behavior and I/O, and augment it to add the capacity to reason about runtime as well. As a capstone proof of concept, we apply the new metrics machinery to an IoT lightbulb controller (already verified with respect to the previous framework) and produce a new specification with time bounds based on arrival of network packets.

Graph Neural Networks for City Policy Recommendations as a Link Prediction Task

Thu, 01 May 2025 00:00:00 GMT

Graph Neural Networks for City Policy Recommendations as a Link Prediction Task Rozario, Consecrata Maria Graph Neural Networks (GNNs) have become a widely utilized tool in recommender systems in various contexts. While recommendation tasks can be approached using a multitude of data structures and types, graph-structured data is particularly well-suited for this domain, as graphs naturally capture a variety of relationships and interactions between entities. By leveraging graph representation learning, we can effectively encode these complex dependencies, enabling robust and context-aware recommendations. We use this methodology in the domain of policy recommendations for urban centers. To recommend policies, we would learn the complex local and global relationships between cities, their environmental features, and currently implemented policies. We construct a graph structure relating cities, implemented policies, and city features, and formulate the policy recommendation task as a GNN link prediction problem, demonstrating its potential to scale data-driven urban governance.

Automatic Detection of Landmark Acoustic Cues in Human Speech

Thu, 01 May 2025 00:00:00 GMT

Automatic Detection of Landmark Acoustic Cues in Human Speech Park, Janette H. This study presents a framework for the automatic detection of the eight landmark acoustic cues in human speech. Landmarks are key articulatory events, produced as a result of minimal vocal tract constriction (e.g., vowels and glides) or closures and releases in the oral region (e.g., nasal, fricative, and stop consonants). A complete landmark detection system is a key step towards an overarching speech analysis system that relies on lexical acoustic cues, as landmarks guide the identification of other acoustic cues in speech. In the proposed framework, the acoustic properties of each of the eight landmark cues are modeled by extracting speech-related measurements and training Gaussian Mixture Models (GMMs). To remove the effects of speaker variability and different recording environments, methods for normalizing speech-related measurements are proposed and evaluated. For a new speech signal, the normalized speech-related measurements are extracted at each time frame and evaluated against the eight trained GMMs to compute the likelihood of each landmark. Using Bayes’ Theorem, the posterior probabilities are calculated to determine the most probable landmark (or absence thereof) at each time frame. The system’s performance is evaluated by comparing the detected landmarks to the manually labeled ground truth landmark annotations.

Single-Cell Language Model for Transcriptomics & Cell Type Annotation

Thu, 01 May 2025 00:00:00 GMT

Single-Cell Language Model for Transcriptomics & Cell Type Annotation Lin, Vincent As single-cell transcriptomics datasets continue to grow in size and biological complexity, current models for cell type annotation remain limited in their generalizability and are often evaluated on only a small fraction of the standardized cell types defined in modern ontologies. Current state-of-the-art models for transcriptomic representation demonstrate that deep learning models can extract rich features on single-cell data but are evaluated on very few cell types and perform poorly on broader datasets. This work introduces a multimodal model architecture that integrates large language models (LLMs) with gene expression encoders to address this scalability gap in cell type annotation. Inspired by vision-language frameworks, our architecture combines a pretrained scRNA encoder with a Perceiver Resampler that maps gene expression profiles into the latent space of a large language model. We construct structured, ontology-grounded datasets of up to 197 cell types and evaluate our model's performance using instruction fine-tuning. Our experiments analyze the impact of integrating language modeling components with scRNA encoders and their benefit on cell type annotation performance for large, diverse datasets. Our results show that while a scRNA encoder may be sufficient for small datasets, our single-cell model leveraging LLMs consistently outperforms the scRNA encoder baseline on larger datasets, with a widening gap in classification performance as data complexity increases, demonstrating the scalability and improved generalizability of our multimodal architecture. We also provide further analysis of the tradeoffs associated with using the natural language domain for biological analysis.

Inference Time Search for Protein Structure Prediction

Thu, 01 May 2025 00:00:00 GMT

Inference Time Search for Protein Structure Prediction Qi, Richard Scaling inference-time compute for deep learning models has led to superhuman performance in games and enhanced reasoning capabilities for language models. However, similar gains have not yet been made in the field of biomolecular structure prediction. We introduce a new paradigm for inference-time search by adding architectural components and a finetuning procedure to state-of-the-art structure prediction models that give rise to a discrete latent space. We implement algorithms for searching and sampling in this discrete latent space and conduct experiments on a small model, demonstrating an increase in oracle and top-1-selected accuracy for predicted protein-protein complex structures.

Designing a Localized Lower Body Negative Pressure Garment for Long-Duration Spaceflight

Thu, 01 May 2025 00:00:00 GMT

Designing a Localized Lower Body Negative Pressure Garment for Long-Duration Spaceflight Chu, Kaitlyn A. Lower Body Negative Pressure (LBNP) has long been explored as a countermeasure to the physiological deconditioning and orthostatic intolerance associated with prolonged microgravity exposure. Traditional LBNP systems, however, are large, stationary devices that require astronauts to remain immobile during use, limiting their integration into daily spaceflight routines. Although more mobile LBNP solutions have emerged, they remain cumbersome and uncomfortable, ultimately still restricting multitasking and reducing operational feasibility. This study introduces the Soft Kinetics INterface (S.K.I.N.), a flexible, wearable structure designed to support the application of localized LBNP. The goal was to evaluate whether targeted negative pressure applied through the S.K.I.N. could replicate the fluid shift effects of a traditional LBNP chamber while improving comfort, mobility, and time-efficiency. The human thigh was chosen as the focus of this technology demonstration due to its known responsiveness to LBNP and its suitability for small-scale implementation. The development of the S.K.I.N. began with finite element modeling (FEM) to identify optimal material properties and structural geometry. Iterative physical prototyping resulted in a sinusoidal silicone waveform design, selected for its mechanical stability and user comfort. The final prototype was then evaluated in three experimental phases: (1) mechanical testing using pressure-sensitive film to assess structural integrity under vacuum, (2) an ex-vivo pig leg study to validate experimental protocols and assess the S.K.I.N.’s ability to induce fluid shifts, and (3) a human study (n=10) comparing fluid shifts between the S.K.I.N. and a scaled-down version of the traditional LBNP chamber. On average, results from the human study showed that the S.K.I.N. successfully induced localized fluid shifts similar to those of the chamber. However, response magnitude varied considerably across participants. Most of the observed effect was driven by female participants, who exhibited more pronounced fluid shifts, while most male participants showed minimal or no measurable response. FEM simulations supported this finding, suggesting that higher fat-tomuscle ratios — more common in women — may enhance tissue deformability and volume displacement, thereby facilitating greater fluid shifts under negative pressure. Although these differences limit generalizability, they also highlight the potential for the S.K.I.N. to serve as a more targeted countermeasure for specific physiologies or user groups. Although the current S.K.I.N. design’s limited surface area constrains its overall effect, the concept shows promise. The ability to deliver targeted fluid shifts in a more mobile, comfortable format could enable integration into dynamic operational settings. Future work should focus on expanding the system to cover larger areas, such as a whole-pants version, and incorporating a portable vacuum source for mobility in both spaceflight and terrestrial applications. Larger, more diverse participant cohorts will also be necessary to assess long-term usability, efficacy, and individual variability in response.

Mantis: A Screen Magnification Tool for Diagram Traversal

Thu, 01 May 2025 00:00:00 GMT

Mantis: A Screen Magnification Tool for Diagram Traversal Patterson, Lydia J. Complex diagrams and charts can be difficult for people who use screen magnification to navigate. A sense of spatial context and of the diagram’s overall structure is oftentimes lost, as magnifiers can only magnify a fraction of the screen at any given time. So, while sighted users have both clarity and full context simultaneously, screen magnifier users often have to choose or split their attention between the two. Existing screen magnifiers are content-agnostic, so the current way of navigating visualizations is freeform and unguided. The burden of figuring out where to explore while retaining a mental model of the diagram is placed entirely on the user. In this paper, we present Mantis—six prototypes of an automatic, content-aware screen magnification tool designed to aid people who have low vision in the traversal of diagrams. Each design experiments with what sorts of information might be provided to help the user retain a sense of context. Further, they each explore how such a tool might use its knowledge of the diagram’s semantic structure to streamline traversal to and from areas of interest to the user. To this end, we evaluate how these proof-of-concepts improve the user’s navigational experience and reduce the user’s cognitive load.

Teacher-Centered Design in Educational Games: Iterative Improvements to the Tragedy of the Commons pSim Dashboard

Thu, 01 May 2025 00:00:00 GMT

Teacher-Centered Design in Educational Games: Iterative Improvements to the Tragedy of the Commons pSim Dashboard Luong, Jacky K. Teaching tools such as the Tragedy of the Commons (ToC) participatory simulation, developed by MIT STEP Lab, have the potential to develop different skills or knowledge compared to single-player educational games. ToC illustrates the challenges of managing shared resources, but its existing teacher dashboard may not be well-suited to support its growing use across various classrooms. Through surveying and interviewing educators along with observing classroom usage, the software's shortcomings and opportunities for improvement were identified. This resulted in the design and implementation of a redesigned teacher dashboard, including a new “central bank” feature that provides structure to support more complex simulations. Additional enhancements improved usability and performance. Evaluations with teachers and controlled playtests demonstrated that these changes show promise in enabling richer classroom dynamics and making facilitation easier. The findings underscore the importance of teacher experience in educational game design.

All Therapies Are Equal - Unless You’re a Bot: Evaluating the Effectiveness of Four Therapy Schools for AI Chatbot Therapists

Thu, 01 May 2025 00:00:00 GMT

All Therapies Are Equal - Unless You’re a Bot: Evaluating the Effectiveness of Four Therapy Schools for AI Chatbot Therapists Liu, Andi This thesis tests two design questions for Large Language Model (LLM) Chatbot Therapists: Which therapeutic school suits an LLM best, and does an explicit Theory-of-Mind (ToM) reflection improve outcomes? We prompted GPT-4.1-mini to act as eight therapists — CBT, Narrative, Psychodynamic, and SFBT, each with and without a ToM step — and held 240 simulated sessions with scripted AI patients. SFBT achieved the greatest projected PHQ-9 improvement (around 4 points), significantly higher than CBT, Narrative, or Psychodynamic approaches. Immediate distress (SUDS) fell modestly and uniformly across schools. ToM reasoning did not alter either measure. The findings show that extra “thinking time” might not automatically translate into therapeutic gain, but also highlight a current strength of LLMs: executing brief, rule-based therapies.

Automated Fiber Coupling with Actuated Mirrors

Thu, 01 May 2025 00:00:00 GMT

Automated Fiber Coupling with Actuated Mirrors Vel, Vetri Senthil Almost all atomic physics experiments rely on precise alignment of lasers. For example, optical fields are used to cool, control, and image atoms in neutral atom arrays. In this thesis, we present a design for mirrors actuated by servos that allow the precise, repeatable alignment of lasers in free space optical setups. We then apply these actuated mirrors to automate fiber coupling, where laser beams are coupled from free space into a fiber waveguide. We present the theory of fiber coupling and use experimental data on the fiber coupling landscape to develop an accurate digital twin. Insights from the combination of the digital twin and experimental data are used to develop a fast and effective algorithm for automated fiber coupling.

ACED: Automatic Concourse Event Detection

Thu, 01 May 2025 00:00:00 GMT

ACED: Automatic Concourse Event Detection Wagner, Luke A. Fans of the San Antonio Spurs often face long delays when traversing the arena or waiting for food. Automatic Concourse Event Detection (ACED) is a novel system designed for tracking these statistics in the Spurs’ arena in real time. We use existing machine learning models and introduce novel processing algorithms to identify the total number of people in each section throughout the arena in addition to tracking the wait times for different restaurants and restrooms. ACED collects and stores this information in a database, which could be used to present fans with up-to-date arena information in a live dashboard to assist them in their in-game decision making. This would improve the overall fan experience, which could encourage fans to buy tickets more frequently. We provide the San Antonio Spurs with a completed implementation of ACED, which is ready to be deployed within the Spurs’ arena.

Ultraviolet-C Powered Air Purifying Respirator (UVC PAPR)

Thu, 01 May 2025 00:00:00 GMT

Ultraviolet-C Powered Air Purifying Respirator (UVC PAPR) Seeyave, Evan The global challenge posed by pandemics, notably COVID-19, has underscored the critical need for advanced personal protective equipment (PPE). This thesis details the development and evaluation of a multi-stage powered air-purifying respirator (PAPR) incorporating direct ultraviolet-C (UVC) germicidal irradiation. The proposed PAPR aims to provide enhanced protection by actively sterilizing air through this UVC chamber immediately prior to inhalation. This approach offers an advantage over traditional filter-based PAPRs by removing both the need to replace filters and pull air with high-power motors, while still neutralizing a broad spectrum of airborne pathogens, including viruses and bacteria. The primary objective of this research is to design, construct, and test a PAPR prototype capable of achieving a high inactivation rate (target 99.9%), thereby offering a robust solution for individuals in high-exposure environments. In addition to the UVC chamber, we also built an alternate ultraviolet-A (UVA) activated titanium dioxide (TiO2) photocatalytic oxidation (PCO) chamber. This work encompasses the overall design of the system, safety considerations, and testing to quantify its pathogen inactivation efficacy and to characterize system performance.

GridFix: A Desktop Application for the Correction of Algorithmically-Generated Beatgrids for Music

Thu, 01 May 2025 00:00:00 GMT

GridFix: A Desktop Application for the Correction of Algorithmically-Generated Beatgrids for Music Shi, Iris Beatgridding is a technique meant to aid DJs in aligning the beats of two different songs. By overlaying a grid of beat markers (a “beatgrid”) on top of a waveform representation of the track being beatgridded, a song’s beats can be visualized and thus easily matched to another’s. State-of-the-art DJ software—like rekordbox by the company AlphaTheta—will algorithmically generate beatgrids for songs. However, these beatgrids are not always accurate and can often be difficult to correct with only the software-provided tools. GridFix is a desktop application designed to be an auxiliary tool for rekordbox, allowing users to correct rekordbox-generated beatgrids by providing additional functionality that rekordbox does not. GridFix’s main advantage is its ability to let users make local changes to small, isolated sections of a beatgrid, a task that is quite hard to achieve in rekordbox. GridFix is fully compatible with rekordbox and fairly easy to learn how to use, as shown by user testing.

Graph Metrics for Improving Cybersecurity on Software Dependency Networks

Thu, 01 May 2025 00:00:00 GMT

Graph Metrics for Improving Cybersecurity on Software Dependency Networks Yao, Darren Z. Modern software ecosystems are deeply interconnected, allowing a vulnerability in a single component to propagate and affect many others. In this thesis, we model software ecosystems as directed graphs, and apply various graph-theoretic metrics to quantify security risk. We compare two deep learning frameworks (PyTorch and TensorFlow) with two traditional software frameworks (npm and PyPI), identifying critical properties of their dependency structures, which motivates several recommendations for improving software supply chain security.

Planning Robotic Cutting Operations

Thu, 01 May 2025 00:00:00 GMT

Planning Robotic Cutting Operations Lunawat, Tarang Classical planning and most PDDL variants operate on the assumption that the number and types of objects present in the environment are known at the time of initialization and neither can nor do change during plan execution. However, there are many domains in which it is helpful and necessary to be able to capture action (or environment) effects that are able to change the existence of objects rather than just facts about these objects. PDDLStream already provides a framework for "certifying" new facts about the environment as necessary throughout plan execution; I propose using PDDLStream to construct a principled way to reason over not just added facts, but also added or removed objects in the environment. In order to do this, I will work within the domain of cutting operations in the kitchen, as this is a domain that both necessitates a lot of object change as objects are cut and often requires chains of these generated objects to be fully reasoned over. Additionally, I will lay the groundwork to use this principled way to reason over new objects to implement different types of cutting operations in the kitchen, with the eventual goal of a robot planner being able to sequence different provided actions to more efficiently work with knives in the kitchen in a human-like manner.

Adaptive Wavefront Estimation Algorithms for High-Contrast Imaging of Exoplanets

Thu, 01 May 2025 00:00:00 GMT

Adaptive Wavefront Estimation Algorithms for High-Contrast Imaging of Exoplanets Manojkumar, Saikrishna The direct imaging of exoplanets orbiting stars outside our solar system remains one of the crucial tools we have available to answer whether there exists life beyond Earth. The light from an Earth-like exoplanet is approximately ten orders of magnitude dimmer than its host star and hence the imaging system of the telescope observing the exoplanet must be able to suppress the starlight to achieve a “contrast” of 10−10 in the image. This is typically achieved using a coronagraph, which blocks the light from the star while allowing the light from the planet to pass through. However, some starlight that leaks through the coronagraph needs to be further removed in the search region for the exoplanet; this region is referred to as the dark hole or dark zone (DZ). Creating a DZ requires the use of focal plane wavefront sensing and control techniques, which estimates the electric field of the starlight in the focal plane of the telescope using a camera and then informs the deformable mirrors (DMs) located upstream of the coronagraph to null these electric fields. Once the DZ is created with a desired contrast, there are still slow, high-order drifts in the optical system that cause the contrast to degrade over the long observation times of the science target. High-order wavefront sensing and control (HOWFSC) techniques are required to maintain the contrast in the DZ while observing a science target. Dark Zone Maintenance (DZM) is a technique that has demonstrated the ability to maintain the contrast in the DZ over long observation times. This algorithm utilizes an Extended Kalman Filter (EKF) to estimate the open-loop electric field at every pixel in the DZ and use this information to inform the control algorithm. The achievable contrast and contrast stability of DZM are determined by several key parameters: the optical system’s drift rate, the photon flux and associated shot noise in the measurement images, and the probe magnitude applied to the DMs for the estimation algorithm. This work quantifies the impact of the drift rate, photon rate, and probe magnitude on the performance of DZM by performing a parameter scan on high-contrast imaging testbeds. The parameter scan was performed on both the in-air High-contrast imager for Complex Aperture Telescopes (HiCAT) testbed at the Space Telescope Science Institute (STScI) and the in-vacuum Decadal Survey Testbed (DST) at the Jet Propulsion Laboratory (JPL). The parameter scan was run in both simulation and on the physical testbed using the contrast in the DZ as a performance metric, and evaluated relative to the photon-noise theoretical bounds to assess the efficacy of the DZM algorithm. The substantial difference between the theoretical bounds and experimental results, on average 70 times worse on HiCAT, motivated the development and implementation of a new DZM algorithm that utilized a separate EKF to estimate the modes of wavefront error derived from the DMs and use that information to correct for the aberrations. This new modal EKF algorithm was tested with a similar parameter scan on the HiCAT simulator demonstrating a nearly 5 times level of improvement relative to the original DZM algorithm simulation performance. The results of this work will inform the design of future algorithms to maintain high contrast during observations for upcoming space telescope missions such as the Habitable Worlds Observatory (HWO).

Incentivizing Data Contributions in Decentralized Collaborative Learning

Thu, 01 May 2025 00:00:00 GMT

Incentivizing Data Contributions in Decentralized Collaborative Learning Wang, Yuxiao In a collaborative learning scheme such as the federated learning model, each user benefits from the data contribution of others. Previous work shows that the federated learning protocol can incentivize users to contribute more than in the competitive equilibrium by penalizing deviations. However, a central controller with access to all the data may raise privacy concerns. In this work, we construct a decentralized collaborative protocol in which users share data without relying on a centralized controller. We then extend this protocol to a repeated game and analyze the competitive equilibrium behavior, along with strategies users can implement to foster collaboration in the repeated setting of the protocol. We provide a quantitative analysis of free-rider behavior under decentralized protocols and compare the amount of information collected with decentralized protocols against that in the centralized protocol.

DisViz: Visualizing real-world distributed system logs with space time diagrams

Thu, 01 May 2025 00:00:00 GMT

DisViz: Visualizing real-world distributed system logs with space time diagrams McMenamy, Josiah This thesis aims to provide an intuitive debugging and learning tool for distributed systems that communicate by message passing. Understanding and debugging distributed systems can be challenging and slow to iterate on, so there is a need for tools that can speed up the time it takes to diagnose the root cause of a bug. There exists significant prior work in creating tools that can aid in the visualization and debugging of distributed system executions, such as the ShiViz log visualizer [13]. This work builds on top of these tools to provide more debugging information, handle large log files, and be easily instrumented in existing systems. We demonstrate using the tool to debug issues in an implementation of the Raft consensus algorithm [34].

Casting Protein Structure Predictors as Energy-Based Models for Binder Design and Scoring

Thu, 01 May 2025 00:00:00 GMT

Casting Protein Structure Predictors as Energy-Based Models for Binder Design and Scoring Nori, Divya Protein binder design has been transformed by hallucination-based methods that optimize structure prediction confidence metrics, such as the interface predicted TM-score (ipTM), via backpropagation. However, these metrics are imperfect proxies for binding affinity and do not reflect the statistical likelihood of a binder–target complex under the learned distribution. In this work, we propose a principled alternative: an energy-based framework that directly extracts the statistical likelihood of a predicted binder–target complex from a structure predictor’s internal confidence distributions. Building on the Joint Energy-based Modeling (JEM) framework, we introduce pTMEnergy, a statistical energy function over structures that is derived from predicted inter-residue error distributions. We incorporate pTMEnergy into BindEnergyCraft (BECraft), a hallucination-based binder design pipeline that maintains the same optimization framework as BindCraft but replaces ipTM with our energy-based objective. Across a diverse panel of challenging protein targets, BECraft achieves higher in silico success rates compared to BindCraft, RFDiffusion, and ESM3. Beyond design, we evaluate pTMEnergy as an unsupervised scoring function for retrospective virtual screening tasks. Without any task-specific supervision or retraining, pTMEnergy consistently outperforms baseline methods across both protein–protein and protein–RNA interaction benchmarks. Our results demonstrate that confidence-derived energy functions offer a powerful and generalizable signal for binder design and scoring.

Efficient Modeling, Optimization, and LLM-Assisted Decision Support for Geothermal Well Arrays

Thu, 01 May 2025 00:00:00 GMT

Efficient Modeling, Optimization, and LLM-Assisted Decision Support for Geothermal Well Arrays Ouko, Edwin O. Geothermal well arrays, which organize multiple geothermal wells into carefully planned geometric configurations, provide an opportunity to enhance energy production capacity and increase fault tolerance of geothermal systems. Closed-loop geothermal systems (CLGS), a type of geothermal well design, promises to allow harnessing of geothermal energy in any location with minimal adverse environmental impact. I demonstrate how the development of these emerging geothermal technologies could be accelerated by recent advances in large language models (LLMs) in conjunction with high-level high-performance programming languages like Julia. In particular, I focus on how LLMs could be used in design brainstorming and to increase efficiency in numerical modeling. I assess the potential of state-of-the-art LLMs such as ChatGPT, Gemini, Claude, Grok, and a domain-specific model, AskGDR, as expert assistants in geothermal research. Owing to the unpredictable reliability of LLMs, there is a constant need for objective evaluation benchmarks in various domains. I propose a novel approach, leveraging Google’s recently introduced AI tool, NotebookLM, to accelerate the generation of quantitative geothermal benchmarks with only new unpublished questions. In addition, I propose the use of blackbox optimization as a computationally less costly alternative to approximate the optimal configuration of CLGS wells in a geothermal array to minimize thermal interference and improve heat energy production. I evaluate several optimization strategies such as Bayesian optimization, particle swarm optimization, natural evolution strategies, differential evolution optimization, Nelder-Mead, and simulated annealing on various performance characteristics such as convergence speed and highest production capacity attained.

A Transformer-Based Foundation Model for Human Microbiome Analysis

Thu, 01 May 2025 00:00:00 GMT

A Transformer-Based Foundation Model for Human Microbiome Analysis Medearis, Nicholas A. The human microbiome plays a crucial role in maintaining our health. Alterations in the microbiome have been linked to various chronic conditions like autoimmune disorders, metabolic diseases, and cancer. While various tools have been developed to study the microbiome, each tool tends to be specialized for a specific task. To overcome this limitation, we report on the development of a foundation model pretrained on 13,524 human microbiome metagenomic samples. The model was then fine-tuned to predict the clinical status of the host. Our model was able to differentiate between healthy and diseased samples in 10-fold cross-validation on the training dataset with an accuracy of 83.7%. On an external validation dataset of 927 samples, our model had an accuracy of 74.9%. Notably, our model performed even better at differentiating diseases from one another. On the diseased samples in the training dataset, it classified samples with an accuracy of 93.3% in 10-fold cross-validation. Together, our results show that generative AI has the potential to transform microbiome research and advance personalized medicine.

Towards an Augmented Reality-based Cyber-Physical Production System Planner

Thu, 01 May 2025 00:00:00 GMT

Towards an Augmented Reality-based Cyber-Physical Production System Planner Mueller, David Investment in automation by small and medium-sized enterprise (SME) manufacturers in the United States has lagged behind their larger counterparts for decades, despite comprising a majority of the nation’s manufacturing industry. The cyber-physical production systems (CPPSs) introduced by Industry 4.0 promise to bolster productivity and efficiency, but only for those enterprises which invest in constituent technologies. These technologies are not easily integrated in existing factories, typically requiring installation of invasive infrastructure and continuous technical support. Robotic integration is typically performed by specialized third-party firms or by in-house staff with extensive technical training, such as engineers. SME manufacturers are particularly sensitive to the complexities of robot integration due to limited access to technologists, and their need for frequent reconfiguration under economies of scope. This thesis introduces Marve: the Mobile Augmented Reality Visual Editor. Marve is a proof-of-concept Android application that enables line workers to directly configure and control an autonomous mobile robot (AMR)-backed hybrid intralogistics system using lowcost consumer hardware. Workers can use Marve’s augmented reality (AR)-based interface to define and visualize the essential geometry and components of such a system. Once configured, workers are able to simulate how the system would respond to their requests to move material throughout the factory. The use of AR enables extensive work to be done at the planning stage of CPPS integration by line workers themselves, bypassing the need for modeling by engineers. Marve relies exclusively on fiducials and visual-inertial odometry (VIO) for localization, and fiducial tags for object tracking, thus eliminating the need for supporting infrastructure. Taken together, these features make Marve an easy on-ramp for SMEs seeking to transition legacy production lines into the CPPSs of Industry 4.0.

Enabling Efficient ML Inference in SigmaOS with Model-Aware Scheduling

Thu, 01 May 2025 00:00:00 GMT

Enabling Efficient ML Inference in SigmaOS with Model-Aware Scheduling Liu, Katie Machine learning inference in multi-tenant cloud environments leads to significant challenges when it comes to minimizing latency and resource contention, especially as models grow in size and complexity. This thesis addresses the cold start overhead and scheduling inefficiencies of multi-tenant ML serving by integrating the RayServe distributed model-serving framework into σOS, a cloud operating system that unifies container and serverless paradigms. The thesis also proposes two model-aware schedulers within σOS that intelligently routes inference requests to reduce the number of cold starts: Model Colocation, which prioritizes placing requests on machines where the required model is already loaded, and Centralized Model Registry, which tracks globally available models to inform scheduling decisions. These policies proactively reduce model load times by reusing cached models. Experimental results on language translation workloads in an 8-node cluster show that these schedulers achieve a ≈ 50% reduction in average inference latency and eliminates roughly 4–5 cold starts per workload, compared to σOS’s default scheduler. Through this model-aware approach to scheduling, our work enables more efficient, scalable, and low-latency ML inference serving in multi-tenant cloud settings.

Impact of efficiency-driven aircraft technology improvements on climate and air quality

Thu, 01 May 2025 00:00:00 GMT

Impact of efficiency-driven aircraft technology improvements on climate and air quality Shukla, Aditeya The impacts of commercial aviation on global climate and air quality have led to an industry-wide movement to reduce its environmental impact. While technological developments in aircraft propulsion, materials, and aerodynamics aim to reduce fuel consumption and CO₂ emissions, these efforts often overlook the full climate and air quality impacts of aviation, especially emissions impacts of NOₓ, CO, HC, soot, and contrails. This study assesses the environmental constraints associated with advancements driven by fuel efficiency by modeling aircraft technologies across narrow-body, wide-body, and regional jet categories. By focusing on near-future technology insertions in materials, aerodynamics, and propulsion, we can compute quantifiable environmental metrics such as temperature changes, global warming potentials, and monetized environmental damages. Our modeling shows that certain propulsion technologies — such as increased component polytropic efficiencies or higher allowable turbine-metal temperatures — can reduce fuel consumption by more than 10% under favorable re-optimizations of engine design. However, they often raise engine core pressures or temperatures in ways that increase NOₓ emissions indices by more than 30%. This can lead to worse air quality damages, offsetting some of the CO₂ savings and in some cases result in a 2% increase in environmental damages on a total net present value (NPV) basis. Primary structure material upgrades consistently reduce both fuel burn and NOₓ emissions. These improvements in air quality from reduced NOₓ result in a 10% reduction of the total NPV from environmental impacts. This analysis shows that focusing on fuel efficiency alone can be an incomplete metric towards understanding the environmental impact of an aircraft. By offering a quantitative assessment of how near-future upgrades can affect both climate and air quality, this study also provides guidance on which technology paths are most effective in reducing the overall environmental impact of aviation.

Digital Symbol Digit Test: Multimodal Behavior Detection and Visualization

Thu, 01 May 2025 00:00:00 GMT

Digital Symbol Digit Test: Multimodal Behavior Detection and Visualization Xu, Jessica J. Neurodegenerative diseases, such as Alzheimer’s, impact many people worldwide and currently have no cure, making early detection essential for effective symptom management and intervention. Traditional diagnostic practices often rely on subjective clinical evaluations that can vary between practitioners, highlighting the need for more objective methods. The digital Symbol Digit Test (dSDT), administered via the Cognitive Health App on an iPad and using the ETVision Eye Tracking System, aims to provide an automated, reliable method to analyze patient cognitive function to detect early signs of impairment through capturing handwriting and gaze data. This thesis builds upon previous work by automating the synchronization of these two data modalities, refining definitions of learning behaviors, and developing pipelines for data processing and visualization. By creating a synchronized multimodal dataset, we can visualize participant behavior for more intuitive interpretation and draw meaningful conclusions. These contributions provide an end-to-end framework for analyzing behavior during the cognitive assessment and lay the groundwork for future development of diagnostic models to detect early signs of neurodegenerative diseases.

Location Verification for Spoofing Detection in Non-Terrestrial Networks

Thu, 01 May 2025 00:00:00 GMT

Location Verification for Spoofing Detection in Non-Terrestrial Networks Schatz, Ensign Nathan Caleb Reliable location awareness is essential for the development of new services and applications in non-terrestrial networks (NTN). The ability of malicious users to report false location information poses a significant threat to NTN performance. This threat introduces the need for a flexible and robust location verification system (LVS) that can reliably detect malicious users. This paper proposes a single-satellite LVS based on round-trip time and angle-of-arrival measurements. We characterize several sources of uncertainty unique to the NTN scenario and examine their combined effect on positioning error. To detect spoofing probabilistically, we approximate the likelihood function for the unknown user position using a Gaussian mixture model and employ a likelihood ratio decision rule for location verification. Results display receiver operating characteristic curves to evaluate the LVS performance under various satellite ephemeris error conditions, spoofing distances, number of measurements available to the system, and wireless channel properties. The proposed LVS is shown to reliably detect spoofing among malicious users.

Triangle Splatting

Thu, 01 May 2025 00:00:00 GMT

Triangle Splatting Xu, Daniel We develop a differentiable rendering method for recovering 3D meshes of scenes from 2D images. Unlike existing approaches, our method does not rely on a differentiable renderers and is compatible with any standard mesh rasterizer. To our knowledge, it is the first mesh-based differentiable rendering method that is not reliant the use of visibility masks entirely. Beyond these conceptual advancements, we implemented a set of highly optimized kernels that enable efficient scene representation on a sparse voxel grid, effectively overcoming the cubic scaling bottleneck faced by similar methods. These innovations result in promising performance on unbounded real-world scenes with complex backgrounds.

Adaptive Control Strategies for Mitigating Spaceflight Fluid Shifts Using Lower Body Negative Pressure and Non-Invasive Fluid Shift Sensing

Thu, 01 May 2025 00:00:00 GMT

Adaptive Control Strategies for Mitigating Spaceflight Fluid Shifts Using Lower Body Negative Pressure and Non-Invasive Fluid Shift Sensing Ortiz, Ciarra Celena Entering a microgravity environment induces cephalad fluid shifts that can lead to cardiovascular and renal-hormonal adaptations that can effect astronaut health and performance in space. The current monitoring strategies for fluid shift lack the ability to track regional fluid shift in real-time, which limits countermeasure efficacy. This thesis aims to highlight the investigation and validation of using prototype non-invasive radiofrequency (RF) sensors for regional fluid shift detection. Additionally, the integration of the feedback from these sensors into Lower Body Negative Pressure (LBNP) chambers could allow for the development of an adaptive Lower Body Negative Pressure regulation framework. Coaxial RF sensors were designed and characterized using tissue phantoms, and tested in a human subject study involving controlled LBNP exposure. Reflection coefficients (S₁₁ and S₂₂) were analyzed to detect regional fluid changes in arm and leg tissue. The preliminary results indicated a statistically significant decrease in the arm reflection coefficients (S₁₁) during active LBNP, which is consistent with fluid being pulled towards the lower body. The leg reflection coefficients (S₂₂) were more variable and did not exhibit statistically significant results, suggesting a need for more investigation with placement and sensor sensitivity. This work demonstrates the potential of using wearable RF sensors for non-invasive fluid shift monitoring and lays the foundation for integrating fluid sensor feedback into adaptive LBNP control protocols to improve astronaut health monitoring and countermeasure personalization.

Robust Inference via Optimal Transport Ambiguity Sets

Thu, 01 May 2025 00:00:00 GMT

Robust Inference via Optimal Transport Ambiguity Sets Wang, Zheyu Uncertainty quantification is pivotal for ensuring the safety and reliability of predictive algorithms in high-stakes applications—ranging from cancer diagnosis to autonomous driving. This challenge is exacerbated by distribution shift, in which the true data–generating distribution diverges from the nominal distribution on which our statistical methods were trained. In this thesis, we formalize distribution shifts via ambiguity sets—metric neighborhoods in the space of probability measures defined by distances such as the Wasserstein metric—and demonstrate that leveraging these ambiguity sets endows two widely used statistical algorithms with distributional robustness. The Kalman filter enables accurate, real-time tracking of latent states by assimilating noisy, indirect measurements over time. Its performance relies on precise state-space models for both the evolution dynamics and the observation process. In practice, uncertainties in these models introduce errors that can significantly degrade filter accuracy. Here, we review two robust Kalman-filter variants that explicitly account for such errors via Wasserstein ambiguity sets. Split conformal prediction, hereafter referred to as conformal prediction, offers a powerful framework for quantifying predictive uncertainty by constructing prediction intervals with finite-sample, distribution-free guarantees. Despite its widespread success, ensuring its validity under train-test distribution shifts remains a significant challenge. We model distribution shifts using ambiguity sets defined by two optimal transport-based metrics and propose two robust conformal prediction algorithms that preserves validity under these shifts. First, we consider ambiguity sets defined by a pseudo-divergence derived from the LévyProkhorov (LP) metric, which captures both local and global data perturbations. We provide a self-contained overview of LP ambiguity sets and their connections to widely used metrics such as the Wasserstein and Total Variation distances. We then establish a natural link between conformal prediction and LP ambiguity sets: by propagating the LP ambiguity set through the scoring function, we reduce complex high-dimensional distribution shifts to manageable one-dimensional shifts, enabling exact computation of the worst-case quantile and coverage. Building on this foundation, we develop valid robust conformal prediction intervals under distribution shifts, explicitly relating LP parameters to interval width and confidence levels. Experimental results on real-world datasets demonstrate the effectiveness of the proposed approach. Next, we extend our analysis to robust conformal prediction over Wasserstein-2 ambiguity sets, deriving a theoretical characterization of the worst-case quantile. However, we identify intractability due to the dependence on the shape of the original score CDF and conclude with potential future directions.

Theoretical Limits of Quantum Ranging

Thu, 01 May 2025 00:00:00 GMT

Theoretical Limits of Quantum Ranging Kartal, Bünyamin The ability to determine distances from dedicated measurements, namely active ranging, is crucial in a variety of systems including localization, radar, and lidar. This thesis establishes the quantum limits and determines the quantum advantage provided by single-beam displaced squeezed states in active ranging. Analytical expressions of the quantum Fisher information (QFI) are provided for monochromatic and continuous-mode waves passing through a thermal loss channel with arbitrary loss and noise conditions. The optimal allocation of system resources for performing displacement and squeezing operations is determined. The optimal allocation consists of apportioning all resources to perform either the displacement operation, providing no quantum advantage, or the squeezing operation. Analytical results are examined in optical and microwave regimes. The optimal gain, i.e., the ratio between the QFI obtained by optimal resource allocation and the QFI obtained by performing only the displacement operation, is derived for the optical and microwave regimes. Quantum advantage afforded by the prototypical heterodyne receiver is also investigated. The results of this thesis pave the way for establishing a foundation of active ranging and provide insights for system design employing currently available quantum technologies.

Generalized Policy Learning with Planning

Thu, 01 May 2025 00:00:00 GMT

Generalized Policy Learning with Planning Yang, Ryan P. Generalized policy learning seeks to find policies that solve multiple tasks within a planning domain. We introduce methods to search for policies independently in a domain from empty initialized policies. As an extension, we also propose a problem setting to learn satisficing policies between domains. In an independent domain, we propose a score function to guide the policy search. Our approach, Policy-Guided Planning for Generalized Policy Generation (PG3), evaluates policies based on how well it can be used to plan. Empirically, we show that PG3 allows generalized policy learning to occur more efficiently than other baselines with PDDL-based problems and policies represented as lifted decision lists. Finally, our experiments show that policies independently learned are qualitiatively similar, prompting further investigation on the possibilities of further accelerating the policy search process.

Meta-Learning Exploration Strategies with Decision Transformers

Thu, 01 May 2025 00:00:00 GMT

Meta-Learning Exploration Strategies with Decision Transformers Welch, Ryan The problem of pure exploration in sequential decision-making is to identify strategies for efficiently gathering information to uncover hidden properties of an environment. This challenge arises in many practical domains, including clinical diagnostics, recommender systems, and educational testing, where data collection is costly and the effectiveness of exploration is critical. Efficient exploration in these contexts strongly depends on exploiting underlying structural relationships within the environment. For instance, recognizing that multiple medical tests may provide overlapping information can reduce the number of tests required to make a diagnosis. Existing exploration approaches drawn from reinforcement learning and active hypothesis testing typically rely on heuristic strategies that require explicit prior assumptions about such structural information. However, when this information is unknown, heuristic methods often lead to redundant exploration, significantly limiting their practical utility in high-stakes domains. Furthermore, these existing approaches do not leverage past experience to improve their exploration efficiency over time. To overcome these limitations, we introduce In-Context Pure Exploration (ICPE), a novel meta-learning framework capable of autonomously discovering and exploiting latent environmental structures across related tasks to guide efficient exploration. ICPE leverages the in-context learning and sequence-modeling capabilities of transformers, combined with supervised learning and deep reinforcement learning techniques to learn exploration strategies directly from experience. Through extensive experiments on synthetic and semi-synthetic exploration tasks, we demonstrate that ICPE is able to efficiently explore in deterministic, stochastic and highly structured environments without relying on any explicit inductive biases. Our results highlight the potential of ICPE to enable more practical exploration strategies suitable for real-world decision-making contexts.

WhatWhen2Ask: Cost-Aware LLM Querying for Autonomous Robots in Uncertain Environments

Thu, 01 May 2025 00:00:00 GMT

WhatWhen2Ask: Cost-Aware LLM Querying for Autonomous Robots in Uncertain Environments Thirumalai, Vittal Autonomous agents operating in real-world environments must make decisions under uncertainty, facing challenges such as partial observability, sparse rewards, and long-horizon planning. While reinforcement learning (RL) enables agents to learn from experience, standard policies often struggle to generalize in the presence of ambiguous tasks or incomplete information. Large language models (LLMs) can provide valuable semantic guidance, but their high computational cost and latency make constant querying impractical. This thesis introduces WhatWhen2Ask, a framework for cost-aware, confidence-driven querying of external multimodal large language models (MLLMs). The agent employs a Deep Q-Network (DQN) as its internal action planner, selectively querying open- and closed-source models (BLIP-2 and GPT-4o) in a hierarchical manner when its confidence is low and external guidance is likely to improve performance. Accepted hints are embedded and fused with structured state representations, supported by tailored reward shaping for improved learning in sparse environments. Evaluated in the HomeGrid environment, WhatWhen2Ask improves the success rate from 38% (DQN-only) to 54%, while querying in fewer than 6% of steps. Ablation studies show that semantic hints, confidence-based querying, selective hint filtering, and hierarchical fallback each contribute meaningfully to performance. These results suggest that principled, confidence-aware LLM querying can enhance decision-making in uncertain environments, offering a step toward more efficient and cost-aware language-augmented agents.

Detecting Errors in Financial Data: A Multi-Agent LLM and Synthetic Data Approach

Thu, 01 May 2025 00:00:00 GMT

Detecting Errors in Financial Data: A Multi-Agent LLM and Synthetic Data Approach Liu, Katherine With the high volume of activity flowing through financial institutions, detecting potential errors remains a critical challenge. This paper addresses two key areas where errors may occur: business name registrations and transactions within valid accounts. Traditional string-matching methods struggle to accurately identify incorrectly written business names that closely resemble existing ones, while existing error detection models for transaction data often suffer from class imbalance, leading to reduced performance on minority incorrect transaction cases. To address these issues, this paper proposes two novel approaches. First, a hybrid method integrating multi-agent Large Language Models (LLMs) with existing string-matching techniques enhances the detection of incorrect business names by capturing subtle variations beyond conventional edit-distance metrics, improving the recall from 0.815 for the baseline model to 0.987 using the proposed method. Second, an improved tabular data generation method for credit card transactions is introduced, leveraging LLMs and class balancing to generate high-quality synthetic data. Using this data to train error detection systems results in a decrease of the false negative rate from 23.47% to 12.84%. Together, these methods enhance the performance of error detection systems, enabling financial institutions to enhance the experiences of their clients.

Switching State Space Modeling via Constrained Inference for Clinical Outcome Prediction

Thu, 01 May 2025 00:00:00 GMT

Switching State Space Modeling via Constrained Inference for Clinical Outcome Prediction Su, Arnold C. In clinical settings, timely and accurate prediction of adverse patient outcomes can help guide treatment decisions. While deep learning models such as LSTMs have demonstrated strong predictive performance on multivariate clinical time series, they often lack interpretability. To address this gap, this thesis proposes a framework that combines the predictive strength of neural networks with the interpretability of latent variable models. Specifically, we develop a constrained inference approach to train a switching state space model—an autoregressive hidden Markov model (AR-HMM)—for outcome prediction. Our method leverages knowledge distillation: a high-capacity LSTM "teacher" model is first trained to predict a target clinical outcome of interest, and its predictive behavior is then transferred to an interpretable AR-HMM "student" model through a similarity constraint during inference. We implement a constrained variational inference approach to estimate the parameters of the student model while aligning its latent representations with that of the teacher model’s. We evaluated our approach using two real-world clinical datasets. Our approach demonstrates predictive performance comparable to state-of-the-art deep learning models, while producing interpretable latent trajectories that reflect clinically meaningful patient states.

Duality, Weight Decay, and Metrized Deep Learning

Thu, 01 May 2025 00:00:00 GMT

Duality, Weight Decay, and Metrized Deep Learning Newhouse, Laker The Muon optimizer has shown convincing evidence that it is faster and more scalable than AdamW for deep learning training, setting speed records for training NanoGPT and scaling up to models with 16B parameters. The theory that led to Muon is called metrized deep learning, a method that suggests assigning norms to each part of a neural network. Chapter 1 begins with an accessible explanation of metrized deep learning, including one of its recurring tools: odd polynomial iterations that act directly on singular values. Chapter 2 reviews duality, a way to modify the gradient that seeks to decrease the loss the most while disturbing the model the least. Pedagogically, duality links four popular optimizers—SGD, Adam, Shampoo, and Muon—under a common framework, steepest descent under a norm. Practically, experiments suggest that duality-based optimizers train faster than AdamW and transfer learning rate across width. Chapter 3 develops tools to enforce weight norm constraints during training, conferring provable and upfront Lipschitz guarantees for transformers. We find that optimizer dynamics matter: switching from AdamW to Muon improves standard weight regularization methods—weight decay and spectral normalization—allowing models to reach equal performance with a lower Lipschitz bound. Leveraging that Muon’s update has a fixed spectral norm, we co-design a weight constraint method called spectral cap that improves the Lipschitz vs. performance tradeoff for MLPs and 2M parameter transformers. Our 4-Lipschitz transformer on Shakespeare text reaches validation accuracy 60%. Scaling to 145M parameters, our 600-Lipschitz transformer reaches 21% accuracy on internet text. However, to match the NanoGPT baseline validation accuracy of 39.4%, our Lipschitz upper bound increases to 10^274. Nonetheless, our Lipschitz transformers train without stability measures such as layer norm, QK norm, and tanh logit softcapping.

Modeling Sequence Uncertainty in Comparative Genomics with a Probabilistic DNA Representation

Thu, 01 May 2025 00:00:00 GMT

Modeling Sequence Uncertainty in Comparative Genomics with a Probabilistic DNA Representation Zhao, Sarah Ann Uncertainty in nucleotide sequences is widespread in bioinformatics, arising from somatic mutations, population-level variation, sequencing errors, and ancestral state inference. Yet, standard formats like FASTA encode DNA deterministically using ASCII string characters, omitting this uncertainty and contributing to pervasive reference biases in genomics. Graph pangenomes have recently emerged to address these limitations by representing genetic variation across populations as bidirected graphs. While promising, these approaches are still developing and are not yet fully integrated with widely used linearly-referenced genomic tools and databases. To bridge this gap, I introduce pDNA (probabilistic DNA), a linearly-referenced data structure that encodes nucleotide-level uncertainty in a vector format compatible with traditional genomics workflows. Each position in a pDNA sequence is represented as a 4-dimension probability vector over the four possible DNA nucleotides, inspired by position weight matrices and one-hot encodings. I also introduce pFASTA, a binary file format for efficient storage of pDNA sequences, along with an open-source software package for generating, manipulating, and analyzing these data. This framework enables uncertainty-aware sequence analysis while maintaining compatibility with existing genomics infrastructure. I apply this framework to ancestral sequence reconstruction.

Online Acquisition of Simulatable Rigid Object Models

Thu, 01 May 2025 00:00:00 GMT

Online Acquisition of Simulatable Rigid Object Models Yang, Ethan How can we build a robot that operates autonomously in a home environment over long periods of time? A key requirement is the ability to perceive and understand its surroundings, including the objects it will interact with. This thesis investigates how a robot can reconstruct previously unknown objects and integrate them into a physics simulation for planning. We explore two methods for reconstructing the 3D geometry of objects and test their performance in simulation and in real-world experiments. Our results demonstrate that a learned depth model enables 3D reconstruction of unknown objects and their successful integration into simulation environments. Additionally, we investigate methods for estimating an object’s inertial parameters, using its reconstructed mesh and through manipulation.

Scaling contrastive learning batch size by two orders of magnitude

Thu, 01 May 2025 00:00:00 GMT

Scaling contrastive learning batch size by two orders of magnitude Tian, Betsy Contrastive learning has emerged as a powerful framework for unsupervised representation learning, allowing models to learn by maximizing agreement between related samples and distinguishing dissimilar ones. However, contrastive learning frameworks are fundamentally limited by the number of negative pairs a model can observe, and memory-intensive backbones constrain practical batch sizes. We introduce a three-phase, adapter-augmented training framework that scales contrastive batch sizes by two orders of magnitude – surpassing previous state-of-the-art learners in both accuracy and speed. First, we co-train the backbone and adapter on small batches to establish a strong initialization. Next, we freeze the backbone and train the adapter alone with very large batches, exposing it to an enlarged negative pool. Finally, we transfer large-batch adapter gradients back into the backbone via segmented backpropagation. We evaluate our method on the PlacesAudio dataset and show promising results for boosting retrieval performance at each phase. By exposing the model to substantially more negatives per effective batch, we achieve higher accuracy at a faster speed than optimizer-stepping baselines. Ultimately, this approach that scales batch size by hundreds of times can be integrated into any contrastive learning framework for more robust representation learning and abundant negative sampling.

Towards Fully Automated Volumetric Analysis of Lung Nodules in Computed Tomography

Thu, 01 May 2025 00:00:00 GMT

Towards Fully Automated Volumetric Analysis of Lung Nodules in Computed Tomography Rubel, Evan Early detection of lung cancer significantly improves patient outcomes, and tracking the growth of lung nodules over time is key to understanding their progression and informing future treatment decisions. However, calculating nodule growth in computed tomography (CT) scans remains a highly manual and time-consuming task. In this work, we develop an automated end-to-end pipeline to compute lung nodule growth using state-of-the-art computer vision techniques. While modern advances in deep learning have all but solved many learning tasks in the domain of natural images, biomedical imaging presents unique challenges due to limited data availability, inconsistent annotations, and deployment constraints. We address these challenges by training robust detection and segmentation models using the LUNA16 and LNDb datasets. On the held-out UniToChest dataset, our methods generalize well, attaining a nodule recall of 77.49%, reducing false positives per scan by a factor of 11.3 compared to existing techniques, and achieving a mean nodule-wise Dice score of 0.6453. We then apply our methods to analyze nodule growth in 1,378 patients from the National Lung Screening Trial; we estimate a median nodule volume-doubling time of 791.23 days across all nodules from the patients that do not receive a cancer diagnosis and a median nodule volume-doubling time of 637.38 days across all nodules from the patients that do receive a cancer diagnosis. We also recall 82.20% of radiologist-annotated nodules that are directly associated with a cancer diagnosis and estimate a shorter median nodule volume-doubling time of 370.11 days for these nodules. By automating lung nodule growth quantification, this work lays the foundation for improved screening protocols, personalized treatment planning, and the development of novel imaging biomarkers. To encourage further work in this area, we release our full software pipeline at https://github.com/evanrubel/nodule_volumes.

Design and Analysis of a 80 GHz Hybrid CMOS Dielectric Resonator Oscillator

Thu, 01 May 2025 00:00:00 GMT

Design and Analysis of a 80 GHz Hybrid CMOS Dielectric Resonator Oscillator Louie, Tiffany This work studies a high frequency, low phase noise, hybrid CMOS oscillator based on a cylindrical dielectric resonator coupled directly to an on chip structure. Dielectric resonators (DR) are known for their high quality factor, low cost, and high temperature stability which makes them a desirable frequency selecting element in design for millimeter-wave (mmWave) applications. Current dielectric resonator oscillators (DRO) have proven to be phase stable, but are limited in frequency (< 40Ghz) due to their implementation with discrete components. However, in increasing the operational frequency up to the GHz range, it is possible to reduce size of the DR and place it directly on top of a cmos chip. We demonstrate, using a 22nm FD-SOI process, the design of a 80Ghz DRO with an area of 4mm² and an oscillator power consumption of 1.95mW. The DRO achieves a simulated phase noise of -128 dBc/Hz at 1MHz and -148 dBc/Hz at 10MHz.

LEO: an LLM-Powered EDA Overview

Thu, 01 May 2025 00:00:00 GMT

LEO: an LLM-Powered EDA Overview Zheng, Sophia Computational notebooks impose a linear structure that impedes data analysts’ sensemaking process with overwritten cells, dead-end code, and fragmented logic. This challenge is especially pronounced when analysts either encounter a notebook authored by someone else or revisit a self-authored notebook after significant time has passed. In both cases, understanding the analysis code becomes convoluted and laborious. To address these barriers, we introduce LEO, a computational notebook tool that operationalizes notebook summarization by leveraging large language models to (1) cluster analysis patterns and (2) trace variable use. LEO organizes code into a two-level hierarchy–General Level Sections and Code Level Actions—integrated with in-line textual summaries filtered on the variable-level, further supporting task-driven exploration. We evaluate the system’s effectiveness in a user study with five computational notebook users across two realistic use cases. Participants reported that LEO streamlined code comprehension and navigation of undocumented notebooks by allowing them to query variables and traverse code cells with greater ease.

Articulated 3D Scene Graphs from Egocentric Vision

Thu, 01 May 2025 00:00:00 GMT

Articulated 3D Scene Graphs from Egocentric Vision Yu, Alan Robotic mapping systems typically approach building metric-semantic scene representations from the robot’s own sensors and cameras. However, these “first person” maps inherit the robot’s own limitations due to its embodiment or skillset, which may leave many aspects of the environment unexplored. For example, the robot might not be able to open drawers or access wall cabinets. In this sense, the scene graph is not as complete, and requires a more capable robot to fill in the gaps by remapping. We narrow these blind spots in current methods by leveraging egocentric data captured as a human naturally explores a scene wearing Project Aria glasses, giving a way to directly transfer knowledge about articulation from the human to any deployable robot. We demonstrate that, by using simple heuristics, we can leverage egocentric data to recover models of articulate object parts, with quality comparable to those of state-of-the-art methods based on other input modalities. We also show how to integrate these models into 3D scene graph representations, leading to a better understanding of object dynamics and object-container relationships. We finally demonstrate that these articulated 3D scene graphs enhance a robot’s ability to perform mobile manipulation tasks, showcasing an application where a Boston Dynamics Spot is tasked with retrieving concealed target items, given only the 3D scene graph as input.

Decentralized Declustering of Multiple Underactuated Autonomous Surface Vehicles

Thu, 01 May 2025 00:00:00 GMT

Decentralized Declustering of Multiple Underactuated Autonomous Surface Vehicles Strømstad, Filip Traasdahl Multi-agent systems have seen a significant rise in research interest, enabled by the increasing availability of low-cost autonomous platforms and motivated by a wide range of emerging applications. However, the coordinated deployment of large numbers of autonomous vehicles in marine environments remains a nontrivial and high-risk problem, yet it is often overlooked in the literature. These vehicles are typically deployed from a single location, and their underactuated nature, close proximity, and susceptibility to external disturbances make it difficult to achieve a mission-ready configuration without collisions. In this thesis, we address the problem of transitioning a set of underactuated Autonomous Surface Vehicles (ASVs) from arbitrary and inconvenient initial conditions to a deconflicted set of deployed vehicles. We propose a decentralized and scalable method that calculates and assigns target positions to the vehicles, generates optimal paths that comply with minimum turning radius constraints, and ensures collision avoidance between the vehicles through a shared speed policy. Contributions also include a formal definition and quantification of clustering and declustering in multi-agent systems. The approach is implemented using the MOOS-IvP autonomy framework, and performance is evaluated through simulation with up to $64$ vehicles and extensive field trials with eight vehicles. Results demonstrate that our approach reduces the time to decluster for the most challenging initial conditions by 50% compared to the current manual method. By improving efficiency and robustness while eliminating human involvement, this work streamlines ASV fleet deployments, enabling more scalable multi-agent field operations.

DBOS Advanced Network Analysis Capability for Collaborative Awareness

Thu, 01 May 2025 00:00:00 GMT

DBOS Advanced Network Analysis Capability for Collaborative Awareness Lockton, Sophia E. Collaborative cyber defense is an essential strategy for detecting and mitigating cyber threats [1]. As traditional intrusion detection systems struggle against increasingly sophisticated attacks, we propose embedding collaborative cyber defense directly into system infrastructure. This work presents a novel implementation of collaborative awareness within DBOS (a Database-Oriented Operating System), resulting in a platform that significantly accelerates application development while providing built-in security for transactional web services. By treating security as a first-class operating system service, our approach facilitates real-time comprehensive network observation and analysis without the need for external tools. The implementation supports the construction, aggregation, and analysis of traffic matrices using both Python and PostgreSQL-based workflows. These workflows extract and process IP-level metadata from DBOS applications, enabling multi-instance aggregation and analysis of network data. This integration represents the first instance of collaborative network analysis within an operating system runtime, demonstrating that secure-by-default infrastructure is both feasible and performant.

Minding the Politeness Gap in Cross-cultural Communication

Thu, 01 May 2025 00:00:00 GMT

Minding the Politeness Gap in Cross-cultural Communication Machino, Yuka Misunderstandings in cross-cultural communication often arise from subtle differences in interpretation, but it is unclear whether these differences arise from the literal meanings assigned to words or from more general pragmatic factors such as norms around politeness and brevity. In this paper, we report three experiments examining how speakers of British and American English interpret intensifiers like “quite” and “very,” finding support for a combination of semantic and pragmatic factors. To better understand these differences, we developed a computational cognitive model where listeners recursively reason about speakers who balance informativity, politeness, and utterance cost. A series of model comparisons suggest that cross-cultural differences in intensifier interpretation stem from (1) different literal meanings, (2) different weights on utterance cost. These findings challenge accounts based purely on semantic variation or politeness norms, demonstrating that cross-cultural differences in interpretation emerge from an intricate interplay between the two.

Empirical Analysis of Neural Architectures and Side Information in Financial Time Series Forecasting

Thu, 01 May 2025 00:00:00 GMT

Empirical Analysis of Neural Architectures and Side Information in Financial Time Series Forecasting Senthil, Swathi This thesis investigates the predictive capabilities of neural networks in financial time series forecasting, focusing on predicting the weekly close price of the SPY index. We explore the integration of options-derived features alongside traditional price data, compare recurrent architectures and transformer-based models, and evaluate multiple training strategies. Our key contributions include: (1) evidence that options-derived input features improve both error metrics and directional accuracy; (2) a comparison study of four training methods (one-step-ahead, direct multi-step, simulation error, and teacher-forcing); (3) the development of a bidirectional GRU-LSTM hybrid model that outperforms standard recurrent networks in multi-step forecasting; and (4) a novel coarse tokenization approach for discretizing continuous financial data, which improves first-week prediction performance when used in transformer models that use an asymmetric attention mechanism. Overall, this thesis illustrates the importance of input design, model architecture, and training methodology in neural financial forecasting. We conclude by outlining directions for future work, including cross-asset generalization and further exploration of tokenization schemes for transformer-based models.

Mitigating LLM Hallucination in the Banking Domain

Thu, 01 May 2025 00:00:00 GMT

Mitigating LLM Hallucination in the Banking Domain Sert, Deniz Bilge Large Language Models (LLMs) offer significant potential in the banking sector, particularly for applications such as fraud detection, credit approval, and enhancing customer experience. However, their tendency to "hallucinate"—generating plausible but inaccurate information—poses a critical challenge. This thesis examines existing strategies for mitigating LLM hallucinations and proposes a novel approach to reduce hallucinations in the context of predicting customer churn using LLMs.

Layered Unlearning for Adversarial Relearning

Thu, 01 May 2025 00:00:00 GMT

Layered Unlearning for Adversarial Relearning Qian, Timothy Our goal is to understand how post-training methods, such as fine-tuning, alignment, and unlearning, modify language model behavior and representations. We are particularly interested in the brittle nature of these modifications that makes them easy to bypass through prompt engineering or relearning. Recent results suggest that post-training induces shallow contextdependent “circuits” that suppress specific response patterns. This could be one explanation for the brittleness of post-training. To test this hypothesis, we design an unlearning algorithm, Layered Unlearning (LU), that creates distinct inhibitory mechanisms for a growing subset of the data. By unlearning the first 𝑖 folds while retaining the remaining 𝑘 − 𝑖 at the 𝑖th of 𝑘 stages, LU limits the ability of relearning on a subset of data to recover the full dataset. We evaluate LU through a combination of synthetic and large language model (LLM) experiments. We find that LU improves robustness to adversarial relearning for several different unlearning methods. Our results contribute to the state-of-the-art of machine unlearning and provide insight into the effect of post-training updates.

Mixed-Variable Bayesian Optimization using Prior-Data Fitted Networks

Thu, 01 May 2025 00:00:00 GMT

Mixed-Variable Bayesian Optimization using Prior-Data Fitted Networks Qian, Janet Bayesian optimization (BO) is a powerful framework for optimizing expensive blackbox functions, widely used in domains such as materials science, engineering design, and hyperparameter tuning. Traditional BO relies on Gaussian processes (GPs) as surrogate models, but GPs face limitations in flexibility and scalability. Prior-Data Fitted Networks (PFNs) have recently emerged as a promising alternative, leveraging transformer architectures and in-context learning to approximate posterior predictive distributions (PPDs) in a single forward pass. By training on large amounts of synthetically generated data from sample-able function priors, PFNs can learn to rapidly predict PPDs across a wide range of function classes. In this thesis, we investigate the application of PFNs to mixed-variable BO, a particularly challenging setting due to the interplay between continuous and discrete inputs and the combinatorial complexity of the search space. We evaluate how PFNs perform when integrated with a range of mixed-variable BO strategies, including various encoding schemes and discrete-aware acquisition optimization. Additionally, we explore how finetuning PFNs on targeted function priors can enhance performance when prior knowledge about the objective is available. Our contributions include empirical evaluations of mixed-BO techniques, insights into PFN training, and a suite of mixed-variable benchmark problems.

Eliminating Hallucination-Induced Errors in Code Generation with Functional Clustering

Thu, 01 May 2025 00:00:00 GMT

Eliminating Hallucination-Induced Errors in Code Generation with Functional Clustering Ravuri, Chaitanya Modern code–generation LLMs can already solve a large fraction of programming problems, yet they still hallucinate subtle bugs that make their outputs unsafe for autonomous deployment. We present functional clustering, a black-box wrapper that eliminates nearly all hallucination-induced errors while providing a tunable confidence score. The wrapper samples many candidate programs, executes each on a self-generated test suite, and clusters candidates whose I/O behavior is identical; the empirical mass of the largest cluster serves as an exact confidence estimate. A single scalar threshold on this estimate lets users trade coverage for reliability with exponential guarantees. On LiveCodeBench our verifier preserves baseline pass@1 on solvable tasks yet slashes the error rate of returned answers from ∼65% to 2%, and drives it to 0% at a conservative threshold while still answering 15.6% of prompts. Manual audits show that the few residual mistakes stem from prompt misinterpretation, not random generation noise, narrowing future work to specification clarity. Because the method requires only sampling and sandbox execution, it applies unchanged to closed-source APIs and future models, offering a practical path toward dependable, autonomous code generation.

Choosing Networks for Ride-Hailing Platforms

Thu, 01 May 2025 00:00:00 GMT

Choosing Networks for Ride-Hailing Platforms Somsirivattana, Thana The development of autonomous vehicles is poised to reshape the landscape of transportation. As companies prepare to deploy these vehicles on ride-hailing platforms, a key operational challenge is determining the networks on which to train the vehicles. Our work contributes toward addressing this challenge on three fronts. First, we develop a theoretical model of the network selection problem and prove theoretical results that show the importance of two parameters: the detour factor and the fleet size. Second, we develop several approaches for selecting the networks. Third, we evaluate these approaches on empirical data. We find empirical support for the importance of the detour factor and the fleet size.

PyGridSim: A Functional Interface for Distributed System Simulation

Thu, 01 May 2025 00:00:00 GMT

PyGridSim: A Functional Interface for Distributed System Simulation Zhao, Angela M. This thesis details the development of PyGridSim, an open source python module that leverages OpenDSS capabilities to provide an efficient and scalable functional interface for building distributed system simulations. Distributed power systems encompass all components that power an electrical system— from larger power plants to microgrids—and represent the network of electric consumption and production in a system. Simulations of such power systems allow experts to analyze potential faults and risks in a fast, reproducable, and cost-efficient way. Thus, the accessibility of such simulations is critical to supporting the safety and reliability of power systems. While existing packages built for distributed system simulation provide the necessary computing power and customizability of a distributed system simulator, their interfaces are hard to scale over many nodes and often have difficult-to-learn syntax. PyGridSim aims to build on these existing modules—maintaining customizability while providing a flexible, intuitive, and scalable syntax structure.

Improving the Programmability of A Distributed Hardware Accelerator

Thu, 01 May 2025 00:00:00 GMT

Improving the Programmability of A Distributed Hardware Accelerator Shwatal, Nathan A. Sparse iterative matrix algorithms are critical to many scientific and engineering workloads, yet they perform poorly on conventional hardware. (Ōmeteōtl, a new hardware accelerator with a distributed-memory and task-based execution model, aims to address these performance bottlenecks. However, programming for (Ōmeteōtl is low-level, error-prone, and far removed from the simplicity of typical iterative formulations. This thesis presents Lapis, a domain-specific language and compiler that allows users to express sparse matrix algorithms in high-level Python code and automatically generates efficient C++ code for (Ōmeteōtl. Lapis abstracts away data partitioning and task orchestration, reducing implementation complexity: for example, it lowers lines of code by 30× for conjugate gradients and 46× for power iteration. Despite this abstraction, generated code achieves 75.7% to 92.6% of the performance of manually written implementations across several benchmarks.

Study of Thermochemical Non-equilibrium and Sensor Cavity Geometry in Hypersonic Flow

Thu, 01 May 2025 00:00:00 GMT

Study of Thermochemical Non-equilibrium and Sensor Cavity Geometry in Hypersonic Flow Mao, Grace This work presents a computational investigation of the influence of geometric configurations within a hypersonic flow field on optical distortion, with a particular focus on the effects of window deformation and the role of thermochemical modeling compared to perfect gas assumptions. Turbulent RANS and conjugate heat transfer were used to model three 3D geometries in US3D, an unstructured-grid finite volume computational fluid dynamics (CFD) solver. The three investigated geometries are a flat plate with a flush-mounted sensor, an open cavity with a length-to-depth ratio of 2, and a closed cavity with a length-to-depth ratio of 16. The data demonstrate that the flat plate configuration has the best optical performance and that the closed cavity has the worst. Additionally, the inclusion of thermochemistry in the flow simulation results in a more pessimistic outlook on image quality compared to the perfect gas model. The results document optical distortion for several different geometries with and without thermochemical modeling within hypersonic flow that can inform future design decisions and research.

BlueVeri: Formal Security Verification for Bluespec Processor Designs

Thu, 01 May 2025 00:00:00 GMT

BlueVeri: Formal Security Verification for Bluespec Processor Designs Wang, Shih-Yu There are numerous hardware security defense mechanisms designed to mitigate sidechannel attacks. However, ensuring that a defense can comprehensively protect against an entire class of attacks, while avoiding the introduction of new vulnerabilities that could lead to additional attack surfaces, remains a significant challenge. Although researchers have attempted to apply formal verification techniques to hardware security, these efforts have been hindered by scalability issues. In this paper, we introduce BlueVeri, a systematic and automatable approach for formally verifying the security of a Bluespec processor against speculative execution attacks. BlueVeri leverages the high-level information provided by Bluespec’s guarded atomic actions, simplifying and accelerating the verification process. We evaluate BlueVeri on out-of-order processors implemented in Bluespec, demonstrating that our approach substantially enhances verification scalability and is capable of proving the security properties of a minimal out-of-order processor within one hour.

Stairway to Autonomy: Hierarchical Decision-Making for LLM-Guided Planning, Bandit-Driven Exploration, and Multi-Agent Navigation

Thu, 01 May 2025 00:00:00 GMT

Stairway to Autonomy: Hierarchical Decision-Making for LLM-Guided Planning, Bandit-Driven Exploration, and Multi-Agent Navigation Nayak, Siddharth Nagar Autonomous multi-agent systems must efficiently plan, explore, and navigate in dynamic and unknown environments, particularly for tasks like search & rescue and environmental monitoring. These settings are often characterized by partial observability, limited communication, and dynamic objectives that require flexible coordination across agents. Designing autonomy that scales with team size and task complexity requires modular decision-making systems capable of high-level reasoning, information-driven exploration, and robust decentralized execution. This dissertation presents a hierarchical decision-making framework that addresses these challenges across three complementary levels of autonomy: high-level planning, adaptive exploration, and decentralized scalable navigation. At the highest level, LLaMAR (Language Model-based Long-Horizon Planner for Multi-Agent Robotics) leverages large language models (LLMs) to decompose long-horizon tasks into structured subtasks, enabling agents to adapt their strategies dynamically. However, the effective execution of these plans requires knowledge about the environment. Our mid-level exploration strategy, BaTMaN (Banditbased Tracking and Monitoring and Navigation), systematically prioritizes waypoints that maximize information gain while balancing real-world constraints such as energy efficiency and sensor reliability. Finally, InforMARL provides a scalable, decentralized navigation by leveraging graph-based local information aggregation, improving sample efficiency, and demonstrating transferability to unseen team sizes. This dissertation develops each of these modules to address a distinct level of the autonomy stack. LLaMAR functions as the high-level planner, translating natural language goals into structured sequences of subtasks and incorporating real-time corrections through a plan-act-correct-verify cycle. BaTMaN serves as the mid-level exploration engine, guiding sensor-equipped agents to prioritize informative regions based on uncertainty. InforMARL operates at the execution level, enabling decentralized agents to navigate through dynamic environments using graph-based local information aggregation and reactive control policies. Each module is independently deployable and optimized for different challenges: strategic reasoning, data-efficient monitoring, and scalable navigation, respectively. When combined, the three modules form a coherent autonomy stack for multi-agent systems operating under uncertainty.

Machine Learning Methods for Churn Prediction and Infrastructure Resilience

Thu, 01 May 2025 00:00:00 GMT

Machine Learning Methods for Churn Prediction and Infrastructure Resilience Agrawal, Shreeansh This thesis investigates how advanced machine learning methods can effectively address two critical business challenges facing the telecommunications industry: short-term customer churn prediction and long-term infrastructure resilience to climate-driven disruptions. In the first part of this work, I develop an upgrades-informed churn forecasting model tailored specifically for marketing operations. Recognizing limitations in the existing aggregate forecasting methodologies, I create a cohort-based cascade model that explicitly integrates customer upgrade behavior across various contract tenures. To address data sparsity and longitudinal gaps in newer contract types, I employ synthetic data generation and imputation techniques, such as regression-based methods and Multivariate Imputation by Chained Equations (MICE). For forecasting churn and upgrade rates, I prioritize interpretability by applying linear regression enhanced with time-series forecasting techniques and macroeconomic indicators, including the Consumer Price Index. This approach significantly improves forecasting accuracy, aligns internal stakeholder objectives, and supports strategic decision-making around customer retention and promotional offers. The second part focuses on building predictive models and strategic frameworks for long-term infrastructure resilience in the face of increasing climate risks. Leveraging spatial-temporal clustering methods (DBSCAN) and advanced neural network architectures, I develop a model to attribute historical outages to extreme weather events. Further, I integrate this model with future climate scenarios from CMIP5 projections using Monte Carlo simulations, providing actionable insights into future infrastructure vulnerabilities. Employing SHapley Additive exPlanations (SHAP), I interpret model predictions, highlighting critical factors such as precipitation, windspeed, and atmospheric pressure. Additionally, I propose frameworks for quantifying financial impacts of future outages and recommend optimization strategies for proactive infrastructure hardening and emergency response. Collectively, these applications demonstrate the value of strategically employing interpretable and robust machine learning methodologies to enhance short-term operational decisions and long-term strategic planning within telecom organizations.

Segmentation Based Tracking for Aerial Robot Global Localization in Unstructured Environments with Oblique Monocular Camera Orientation

Thu, 01 May 2025 00:00:00 GMT

Segmentation Based Tracking for Aerial Robot Global Localization in Unstructured Environments with Oblique Monocular Camera Orientation Shafferman, Hannah R. In the field of robotics, there has been a growing interest in multi-robot systems and their potential to improve the efficiency, scale, and reliability of tasks beyond what an individual robot can achieve. Global localization is a crucial task for autonomous robot navigation, specifically in the multi-agent scenario where robots need to localize within maps communicated by other agents. The scenario where vehicles are viewing their environments from the same perspective, or camera viewpoint, is well studied. However, when environments are mapped from different camera viewing angles, traditional methods fail to match visual features and thus fail to localize. The technical gap that this thesis addresses is when autonomous vehicles within a team are mapping the same environment from different viewpoints, specifically nadir and an oblique camera orientations in an unstructured environment. Many existing visual place recognition (VPR) methods fail to match visual features that look visually different due to appearance, illumination, or viewpoint changes and thus fail to localize. In this thesis, we demonstrate the shortcomings of previous work to generalize to an off-nadir camera angle and explore the benefits and challenges that arise with utilizing oblique imagery for visual feature detection and tracking. We propose a segmentation-based object tracking pipeline to improve tracking and environment mapping performance in this traditionally challenging scenario. Our approach consists of 1) a front-end auto-segmentation tracking pipeline followed by 2) a submap correspondence search, which exploits geometric consistencies between environment maps to align vehicle reference frames. We evaluate our approach on a challenging indoor, cluttered dataset and demonstrate a maximum precision 74% higher than traditional and learning-based baseline methods, with a map size 0.5% the size of the most memory conservative traditional baseline method.

An Aerocapture Guidance and Estimation Framework for Improved Robustness to Uncertainty

Thu, 01 May 2025 00:00:00 GMT

An Aerocapture Guidance and Estimation Framework for Improved Robustness to Uncertainty Sonandres, Kyle A. Aerocapture is an orbital insertion maneuver that converts a hyperbolic approach trajectory into a desired captured orbit using the aerodynamic forces generated during a single atmospheric pass. While it offers major benefits, such as reduced interplanetary cruise time and lower propellant mass reserves, it also introduces significant risk due to extreme sensitivity to atmospheric and delivery state uncertainties. This drives the need for robust guidance algorithms and accurate environmental estimation techniques. This thesis presents approaches to address both of these needs, developing solutions to improve aerocapture performance and robustness to uncertainty. The first contribution is the development of ABAMGuid+, a novel aerocapture guidance algorithm that leverages simultaneous control over bank angle and angle of attack. Inspired by optimal control theory, the algorithm uses a four-phase structure to mimic the optimal control laws while maintaining tractability for online use. Optimal control theory is utilized to identify the optimal control solutions, and numerical optimization is used to validate the analytic solutions prior to integration into a guidance algorithm. Extensive simulation results of a Uranus aerocapture scenario, including over 140,000 Monte Carlo trajectories, demonstrate significant improvements in capture success rates and propellant efficiency compared to existing methods. The second contribution addresses environmental uncertainty directly by developing a deep learning-based approach to estimate the atmospheric density profile during flight. A long short-term memory (LSTM) neural network-based architecture is trained to predict atmospheric density given sequences of flight data. The trained model is integrated into the guidance loop and a curriculum learning process is used to refine in-flight performance. Monte Carlo results show that the LSTM-augmented guidance system reduces propellant usage compared to traditional estimation methods. In summary, this thesis presents two approaches that improve aerocapture performance and robustness to uncertainty. We show that this added robustness can be achieved both by expanding algorithmic ability and by improving environmental estimation approaches.

Mass and Distance Estimation Simulations for the Nancy Grace Roman Space Telescope Using PyLIMASS and a Case Study on Intellectual Property Frameworks in Space Collaborations

Thu, 01 May 2025 00:00:00 GMT

Mass and Distance Estimation Simulations for the Nancy Grace Roman Space Telescope Using PyLIMASS and a Case Study on Intellectual Property Frameworks in Space Collaborations McGee, Carissma Gravitational microlensing is a phenomenon in which a foreground star or planet briefly magnifies light from a more distant background star. This effect enables the discovery of exoplanets that are otherwise undetectable, including those orbiting faint hosts and at large separations. Microlensing is well suited to characterizing exoplanets beyond the snow line, revealing mass ratios and orbital geometries inaccessible to transit or radial velocity methods. The Nancy Grace Roman Space Telescope will carry out the Galactic Exoplanet Survey to detect thousands of microlensing events with the cadence and precision necessary for statistical exoplanet population studies. To verify Roman’s ability to meet its core science requirement, recovering the lens mass and distance in at least 40% of planetary events with better than 20% uncertainty, targeted simulations are essential. Using the pyLIMASS inference framework and Fisher matrix-based uncertainty propagation, I demonstrate that for the well-characterized event OGLE-2013-BLG-0132Lb, the lens mass can be constrained to within 18.7% uncertainty, validating the feasibility of Roman’s requirement on a case-study basis. This thesis also addresses the legal and policy foundations needed to ensure global access to these simulation tools. By advancing open-source software models and proposing a space IP framework for equitable knowledge sharing, it supports collaborative scientific infrastructure for future international space missions.

Combined Steam Power Cycle and Turbofan Engine for Improvement in Aviation Climate Impacts

Thu, 01 May 2025 00:00:00 GMT

Combined Steam Power Cycle and Turbofan Engine for Improvement in Aviation Climate Impacts Mueller, Anna Despite significant innovations in aviation technology over the last 70 years resulting in enormous efficiency improvement, the rising demand for air travel means that aviation carbon emissions continue to increase each year. The rate of improvement to aircraft propulsion engines is diminishing and additional improvements often add significant engine cost or weight. With the goal of reducing aviation’s contribution to global climate change, future aircraft engine designers must consider concepts that stray from the traditional turbofan engine. In this thesis, I develop an engine cycle model combining the turbofan engine with a steam power cycle and use the model to explore the benefits of applying this concept to aircraft engines. In order to study the impact to engine performance and emissions from adding a steam cycle, the engine model needs to be capable of representing the water phase changes and the heat exchangers required to drive those phase changes. My contribution is the development of such a model – with special attention to the modeling of water properties and phase change of water – which ties heat exchanger models into an engine thermodynamic model. The engine cycle as well as heat exchanger parameters including water-to-air ratio, combustor exit temperature, overall pressure ratio, and water pressure are varied to explore the impact to overall engine performance, including the impact of the added heat exchanger weight. This thesis covers the development and initial testing of this model, which enables future studies in engines with phase changing heat exchangers or water injection with the goal of assisting the search for the future engine technologies that will reduce harmful impacts of aviation while continuing to allow air travel.

An Aero-Thermo-Chemo-Mechanical Coupling Framework for the Analysis of Hypersonic Ablative Thermal Protection Systems

Thu, 01 May 2025 00:00:00 GMT

An Aero-Thermo-Chemo-Mechanical Coupling Framework for the Analysis of Hypersonic Ablative Thermal Protection Systems Hoss, Summer A. There are countless challenges associated with the accurate modeling of the hypersonic flight of ablative thermal protection systems (TPS): resolving the relevant coupled physical phenomena through multi-physics simulations, the management of the disparate spatiotemporal scales associated with the fluid and solid responses, and establishing a reliable numerical model able to predict the response of ablative materials exposed to extreme gradients—to name a few. The two-way, loosely coupled framework presented in this thesis consists of ΣMIT, a multi-physics computational solid mechanics (CSM) code, coupled with US3D, a hypersonic computational fluid dynamics (CFD) solver, to form a complete aero-thermochemo-mechanical simulation framework. The ΣMIT-US3D coupling framework provides a step towards high-fidelity simulation capabilities for hypersonic vehicles with ablative TPS, establishing a strong foundation for the simulation of fluid-structure interaction (FSI) phenomena and computation of the mechanical response of porous ablators. The requirement of a robust numerical formulation for the solution of hypersonic pyrolysis problems was made apparent when encountering numerical convergence issues with legacy methods, which sparked the development of a robust semi-implicit pyrolysis material model. The so-called Linearized Pyrolysis model employs simplifying assumptions for the energy and mass balance equations and relies upon the time-lagging of chosen terms to achieve linear convergence and robust performance. The performance of the model has been validated against the Ablation Workshop Test Cases and has increased the range of allowable representative hypersonic boundary conditions significantly compared to the legacy approach. Together, the model and the coupling framework are applied to two aero-thermochemo-mechanical analyses contained within the thesis: a spherical-tipped nose cone and the Orion heat shield. Preliminary results identify the decomposition region as a zone in which high von Mises stress tends to occur—care must be taken to ensure that internal and external flight loads do not exceed allowable limits to prevent catastrophic TPS material failure in this region. However, perhaps the most significant insight resulting from the framework relates to the computation of mass fluxes through the porous ablative material, revealing that for an isotropic monolithic heat shield with at a zero angle of attack, pyrolysis gas flow is driven by the pressure gradient applied to the shield such that the flow exits at the edges of the shield rather than from the base.

Aeroverse: Aerospace Education in Extended Reality

Thu, 01 May 2025 00:00:00 GMT

Aeroverse: Aerospace Education in Extended Reality Johnson, Mollie Aerospace education is a continuously evolving field that is increasingly dependent on digital tools. However, it is ambitious to shift the teaching paradigm to accommodate new cutting-edge technologies. Extended reality (XR), which encompasses augmented (AR) and virtual reality (VR), is an example of such technology. In recent years, VR has seen an increase in usage in education as a novel way to provide students with immersive learning experiences, and XR has a long history of use within the working aerospace industry. However, application in the overlap between the two— aerospace engineering education— remains largely unexplored to date. The themes addressed in this thesis are two-fold: first, the goal is to create VR learning modules to supplement the existing aerospace engineering curriculum. Second, the aim is to validate whether VR technology as a teaching medium can improve learning outcomes and student engagement within the MIT AeroAstro department. With these themes in mind, two experiments were conducted to explore this topic. The first experiment presents the design and execution of an experimental course aimed at aerospace engineering students to assess the educational impact of VR. Over the course of this study, ANOVA and Kruskal-Wallis tests found that there was no significant difference (p > 0.05) in performance between the VR and non-VR groups, save for a few exceptional cases. The second experiment details the integration of a single VR module into an existing course in which all students interacted with the VR activity. Students responded positively to this experiment, reporting increased feelings of engagement and a sense that it aligned well with the rest of the course. One-sample Wilcoxon tests reveal that these findings are largely significant (p < 0.05). This thesis advances the work on assessing VR use for aerospace education. The implications of this work may influence the decisions of other educators regarding the adoption of VR technology as supplements to their own teaching methodologies. As a whole, this thesis contributes to the broader conversation on integrating VR into the classroom.

Investigating the Role of Mission Architecture in Crew Socioemotional Health for Mars Exploration

Thu, 01 May 2025 00:00:00 GMT

Investigating the Role of Mission Architecture in Crew Socioemotional Health for Mars Exploration MacRobbie, Madelyn Human space exploration is evolving rapidly, with commercial successes and NASA’s Artemis missions driving rapid growth and innovation. Plans for longer, larger, and more complex missions necessitate development of new mission architectures to sustain the crews needed to support these missions. Larger missions and multi-site architectures have become feasible with advances in commercial launch vehicles, and generate increased safety and redundancy for crewed operations. However, crew dynamics in these mission architectures have yet to be investigated. This thesis investigates the role of mission architecture (specifically single-site versus dual-site configurations) in subgroup formation and the resulting impacts to socioemotional well-being. We first develop a systematic approach for optimizing analog mission design, then apply this to design two analog missions to compare the effects of single-site and dual-site mission architectures on crew dynamics and psychosocial health. Results provide valuable insights for future Mars mission design, where crew structure and psychosocial adaptation are critical to mission success.

Towards a Strong, Human-Compatible Codenames AI Agent

Thu, 01 May 2025 00:00:00 GMT

Towards a Strong, Human-Compatible Codenames AI Agent Zhu, Sebastian Current language models are limited in their ability to solve complex planning and reasoning problems without the aid of search procedures. While a large body of work has developed search procedures tailored to single-turn, single-user natural language interactions, language generation in multi-agent contexts involving multiple users, imperfect information, and partially misaligned objectives remains extremely challenging. We aim to build search procedures that will enable language models to assist with interactive, multi-agent decision-making in a diverse range of contexts. Using the word game Codenames as a benchmark, we will combine game-theoretic planning procedures with basic language model-based scoring methods to create agents that both play strong policies and play well with human policies. This work yields a set of practical text generation procedures, new evaluation benchmarks, and foundational algorithmic improvements in language model search.

An Investigation into Contrail Observability from Different Satellite Platforms

Thu, 01 May 2025 00:00:00 GMT

An Investigation into Contrail Observability from Different Satellite Platforms Euchenhofer, Marlene V. Contrails are line-shaped ice clouds that can form behind aircraft engines and, under certain cold and moist conditions, spread into contrail cirrus that persists for several hours. By adding to the existing cloud cover, contrails can act to either cool or warm, with the latter, on average, being dominant, resulting in an overall warming effect. Although the effective radiative forcing from contrails is inferred to be of the same order of magnitude as that caused by aviation’s CO₂ emissions, large uncertainties remain around specific radiative forcing estimates. Observational studies of contrails, either to support climate impact assessments or operational contrail avoidance strategies, face trade-offs between spatial and temporal resolution. Many recent publications have relied on data from geostationary satellites accepting lower input data resolution in exchange for higher temporal resolution and greater spatial coverage. Limitations of the observability of contrails in the resulting images have not been sufficiently investigated and need to be assessed and quantified. This study aims to leverage the higher spatial resolution of VIIRS satellite imagery to identify potential limitations on contrail observability in lower-resolution GOES ABI imagery. We generate a dataset of human-identified contrails visible in false-color thermal infrared imagery from both GOES ABI and VIIRS for twelve scenes over the contiguous US. Based on this dataset, we investigate the number, cover, and appearance of the observed contrails. We find that GOES ABI does not resolve 80% of all contrails that can be identified in VIIRS imagery and only shows half of the total observed contrail length. Finally, incorporating an existing contrail-flight matching algorithm by Barbosa, we show that VIIRS tends to resolve more younger contrails than GOES ABI. The findings from this study help to bound the validity of current contrail simulations and modeling outputs that estimate contrail cover and occurrence.

MINCE: Dialect-Aware SQL Decomposition for Federated Query Execution

Thu, 01 May 2025 00:00:00 GMT

MINCE: Dialect-Aware SQL Decomposition for Federated Query Execution Zhang, Sophie S. The increasing adoption of specialized database systems has led to the rise of heterogeneous data environments. While having multiple engines in a data infrastructure enables opportunities for workload optimization, SQL dialect incompatibility makes workload migration difficult. To address this challenge, we develop MINCE (Multi-dialect INtegration and Crossengine Execution), a technique that decomposes SQL queries into parts to enable federated execution across engines with differing SQL dialects. MINCE uses a rule-based method to partition a query into executable components that are assigned to different database systems. To evaluate different execution strategies, MINCE further implements a cost model that incorporates both on-engine query execution time and inter-system data transfer overhead. We evaluate MINCE on a TPC-H-based workload augmented with PostgreSQL-specific functions unsupported in Amazon Redshift. Experimental results show that MINCE produces the fastest execution strategy among our baselines for 72.1% of queries using estimated cardinality, achieving a 2× speedup over single-engine baselines. With perfect cardinality information available to our cost model, this value increases to 88.4%, with an average 2.8× speedup. These results demonstrate that our system not only enables more flexible federated query execution, but also reliably identifies performant execution strategies.

New results in canonical polyadic decomposition overfinite fields

Thu, 01 May 2025 00:00:00 GMT

New results in canonical polyadic decomposition overfinite fields Yang, Jason Canonical polyadic decomposition (CPD) consists of expressing a tensor (multidimensional array) as a sum of several rank-1 tensors, each of which is an outer/separable product of vectors. The number of rank-1 tensors used in a CPD is called the rank of the CPD, and the minimum possible rank of a CPD for a given tensor is called the rank of the tensor. CPD is at the core of fast matrix multiplication, a computational problem with widespread implications across several seemingly unrelated problems in computer science. Much recent progress in this field has used randomized heuristic search to find new CPDs, often over a finite field. However, if these techniques fail to find a CPD with low enough rank, they cannot prove that no such CPD exists. Consequently, these methods fail to resolve certain long-standing questions, such as whether the tensor corresponding to 3 × 3 matrix multiplication has rank less than 23. To make progress on these problems, we develop a novel algorithm that preserves exactness, i.e. they can provably verify whether or not a given tensor has a specified rank. Compared to brute force, when searching for a rank-R CPD of a n0 × · · · × nD−1-shaped tensor over a finite field F, where n0 ≥ · · · ≥ nD−1, our algorithm saves a multiplicative factor of roughly |F| R(n0−1)+n0( P d≥1 nd) . Additionally, our algorithm runs in polynomial time. We also find a novel algorithm to search border CPDs, a variant of CPDs that is also important in fast matrix multiplication. Finally, we study the maximum rank problem and give new upper and lower bounds, both for families of tensor shapes and specific shapes. Although our CPD search algorithms are still too slow to resolve the rank of 3 × 3 matrix multiplication, we are able to utilize them in this problem by adding extra search pruners that do not affect exactness or increase asymptotic running time.

Parameter Estimation for Anonymous Hawkes Processes

Thu, 01 May 2025 00:00:00 GMT

Parameter Estimation for Anonymous Hawkes Processes Wang, William Hawkes Processes are self-exciting point processes used to model many real-life networks in which an event from one agent causes the rate at which events occur from related agents to increase, such as in earthquake networks or social media. This project investigates the question of finding the underlying structure of the Hawkes Processes given a history of when events occurred. This problem has been studied extensively in the regime where the event labels are known, and the bulk of the literature involves parameterizing the model and passing it through statistical learning tools. Our proposed work focuses on the the same question in “anonymous" case where labels are not given. In this regime, the lack of information makes many previous approaches intractable and we develop novel non-parametric approaches for solving cases of the structure learning problem in algorithmic and information theoretic settings. Our results show the ability to learn the entire model under mild assumptions in the information theoretic regime, where we have access to an arbitrarily long Anonymous Hawkes Process transcript, whereas when we’re confined to a polynomially lengthed transcript, the situation is considerably more difficult.

Organization Infrastructure for Tokenized Asset Records

Thu, 01 May 2025 00:00:00 GMT

Organization Infrastructure for Tokenized Asset Records Whartenby, Patrick E. The Tokenized Asset Record (TAR) represents a way to connect existing technology related to tokenized assets and asset schemas to real-world documents that validate the existence of an object. Exactly who should manage TARs and the properties of the related organization schemes remains an open question. Answering this question is crucial to furthering the existing digital economy. While existing solutions have sought to expand digital commerce through pioneering digital clearing houses, little work has explored support for other classes of real-world digitized assets with proof of ownership and existence. The research proposed here seeks to answer this question by suggesting possible solutions and developing a framework for uniformly analyzing the proposals. The research proposes and evaluates three models for the management of TARs. The first is a scheme that involves each industry setting up its own TAR database and managing the system independently from other industries. The second proposes hosting all TARs on a single blockchain. The third argues for an off-chain decentralized platform to host all, akin to the Data Spaces proposed by the European Union. The research finds, based on the proposed criteria, that a decentralized off-chain approach best meets the goals of a TAR management framework.

Unveiling Phenotype–Genotype Interplay with Deep Learning Foundation Models for scRNA-seq: A Quantitative Perspective

Thu, 01 May 2025 00:00:00 GMT

Unveiling Phenotype–Genotype Interplay with Deep Learning Foundation Models for scRNA-seq: A Quantitative Perspective Thadawasin, Pakaphol Foundation models have emerged as powerful tools for analyzing single-cell RNA sequencing (scRNA-seq) data, leveraging large-scale pretraining to capture complex gene expression patterns. However, a comprehensive quantitative framework for understanding the interplay between phenotypes and genotypes remains underdeveloped. Such a framework is critical not only for validating model performance but also for uncovering previously unrecognized biological relationships. In this work, we present both traditional and deep learning-based quantitative analysis pipelines for PolyGene [1], a transformer-based scRNA-seq foundation model, aimed at disentangling the complex phenotype–genotype relationship. First, we implement a top-k classification and entropy evaluation pipeline to serve as a primary validation framework. Our results demonstrate that the pretrained PolyGene [1] is robust in top-k classification metrics and provides meaningful insights into the entropy landscape of human cells across different life stages. Second, we propose a novel deep learning gradientbased gene selection method designed to address limitations in traditional feature selection approaches, such as poor scalability and sensitivity to heterogeneity in high-dimensional data. Through empirical evaluations on benchmark scRNA-seq datasets, we show that our method enhances model interpretability and improves downstream performance, offering a more scalable and biologically relevant alternative to existing techniques. Overall, this work introduces a set of quantitative analysis tools that fill a critical gap in evaluating and interpreting scRNA-seq foundation models, contributing to a deeper understanding of the genotype–phenotype interplay through modern deep learning techniques.

Deepfake Face Detection: An Ensemble Framework for Generalized Classification in Biometric Verification Systems

Thu, 01 May 2025 00:00:00 GMT

Deepfake Face Detection: An Ensemble Framework for Generalized Classification in Biometric Verification Systems Zen, Hilary Generation methods for deepfake images have advanced rapidly, and deepfake face images pose a critical security for biometric verification systems. Applications that rely on face recognition to grant access to sensitive data need to maintain high accuracy across a wide variety of deepfake generation methods, including novel and developing types that the application has not previously trained on. Current deepfake detection models achieve nearperfect accuracy on benchmark datasets, but do not perform as well on unseen types of deepfakes that were not part of their training dataset. We propose building an ensemble model with multiple base detectors, each trained on different generation model families to maintain high performance across many deepfake generation methods. Using four base models, including two models with the same architecture and training data, we exhaustively test all possible ensemble models. We find that combining similar base models trained on the same deepfake generation family does not improve performance compared to the individual base models. However, combining base models trained on different deepfake generation families leads to significant increases in accuracy and recall. Our ensemble framework provides a flexible and inexpensive solution in the ever-changing landscape of deepfake generation and security.

When Should Model Updates Propagate?

Thu, 01 May 2025 00:00:00 GMT

When Should Model Updates Propagate? Struckman, Isabella Marguerite AI supply chains rely increasingly on downstream developers adapting pretrained upstream models. When upstream models are retrained with data deletions (which may be prompted by copyright violations, privacy compliance, or removal of illicit content), it’s unclear if all downstream developers must also undergo costly retraining. In this thesis, we investigate the propagation of data deletions through fine-tuned models within a controlled visual classification setting comprising dog-breed and plane-manufacturer recognition tasks. We show that not all model updates propagate equivalently to downstream tasks, and there is a strong relationship between the deleted data’s relationship to the downstream task and its affect on the downstream model. We demonstrate that neither simple performance metrics (accuracy or F1), nor output-level divergences, nor even embedding-based similarity metrics alone adequately predict when a deletion meaningfully impacts downstream tasks. To overcome these limitations, we introduce an information-theoretic metric grounded in Gaussian mixture modeling (GMM) of embedding distributions, capturing deeper representational shifts. Our proposed approach precisely distinguishes when deletions require downstream retraining, achieving high predictive accuracy and recall without directly accessing retrained downstream models.

Digital Twin Modeling for NV Magnetometry

Thu, 01 May 2025 00:00:00 GMT

Digital Twin Modeling for NV Magnetometry Rich, John P. This thesis presents the development and application of a digital twin modeling framework for nitrogen-vacancy (NV) center-based magnetometry, advancing the field of quantum sensing. A surrogate model serves as a computational representation of the physical NV magnetometer system, enabling comprehensive exploration of parameter spaces to optimize device design. Leveraging machine learning techniques, this study optimizes control mechanisms, including the design of learned analog filters, to enhance system performance. This research investigates the fundamental limits of NV magnetometer performance, identifying strategies to minimize power requirements while maintaining high sensitivity. A dynamic framework is implemented to update the surrogate model’s parameters in real-time based on experimental measurements, ensuring accurate fidelity to the physical system. Additionally, the optimized control strategies are simulated within the digital twin environment, demonstrating their potential for advanced quantum sensing applications such as magnetocardiography (MCG) for heartbeat detection.

Fuzzing for User-Schedulable Languages

Thu, 01 May 2025 00:00:00 GMT

Fuzzing for User-Schedulable Languages Moon, Kenneth Performance engineers restructure programs to use hardware as efficiently as possible. Even simple mathematical functions can become sprawling and complex programs when fully optimized, as the resulting code must often be precisely molded around specialized behaviors supported by the hardware. To help performance engineers deal with this complexity, userschedulable languages provide scheduling operations, which are abstractions of common steps taken to restructure programs. By composing these scheduling operations, performance engineers can concisely represent their intended optimizations to programs. Exo, being a user-schedulable language, provides this abstraction with the additional guarantee that any scheduling operation which passes Exo’s automated checks does not change the behavior of the program. Though this guarantee is useful for avoiding bugs while optimizing a program, the analysis required to provide such a guarantee is infeasible on programs in general. To make analysis feasible, Exo only allows users to write programs with a restricted set of behaviors. As a result, some programs are impossible to schedule using Exo, limiting the use cases of Exo. In this thesis, we explore how fuzzing can be used as an alternative to the existing analysis in Exo, with the goal of allowing Exo to analyze more complex programs. “Fuzzing” refers to a test case-driven approach to determining properties of a program, such as whether its behavior changes after a scheduling operation. If the program’s outputs do not change after the scheduling operation when provided the same inputs, the fuzzer concludes that the program’s behavior did not change. Since fuzzing only requires us to know how to evaluate the program, it can be applied to a much broader set of programs than the existing analysis in Exo. However, fuzzing can miss mistakes in scheduling if the fuzzer fails to find a test case demonstrating the issue with a scheduling operation, as it is a complete form of analysis rather than a sound form of analysis like the existing analysis in Exo. Additionally, fuzzing can be costly compared to the original analysis, as repeatedly running programs on many test cases for many scheduling operations can be slow. We explore ways to mitigate these issues throughout this work. Finally, we evaluate our implementation of the fuzzer and its performance on some example use cases for Exo.

Efficient Verifiable Computation Made Easy

Thu, 01 May 2025 00:00:00 GMT

Efficient Verifiable Computation Made Easy Ma, Chengyuan Recent advancements in cloud computing, data privacy, and cryptography have sparked a growing interest in Verifiable Computation (VC) in both industry and academia. In particular, zero-knowledge proof (ZKP) algorithms are gaining rapid traction due to their strong privacy guarantees. However, they are notoriously computationally intensive, making performance a critical concern. Given the inherent data parallelism and heavy use of vector operations in ZKP computations, multicore CPUs and GPUs offer a promising acceleration path. Unfortunately, accelerated programming for ZKP remains challenging: ZKP algorithms evolve rapidly, their structures grow increasingly complex, and writing high-performance ZKP code is tedious, error-prone, non-portable, and unfriendly to algorithm developers. We present an end-to-end compiler framework, Zera, that lowers ZKP algorithms to parallel hardware for efficient acceleration, with minimal programmer effort. By effectively leveraging ZKP algorithm patterns and trends, we are able to automate the key performance optimizations, with a succinct linguistic extension and a set of practical compiler customizations. Consequently, with just 92 lines of trivial high-level annotation added to the original 7,000 lines of C++ code, our single-source code solution delivers 33.9× and 24.0× speedup on GPU over a highly optimized serial C++ implementation on CPU and an existing multithreaded Rust baseline on CPU, respectively. Compared to our hand-optimized GPU/CUDA implementation requiring an extra 2,000 lines of low-level code (roughly 60 programmer hours), our compiler-generated GPU implementation is only 58% slower (1.58× slowdown) on large inputs, demonstrating a compelling trade-off between performance and productivity.

Optimizing Partitioning for Efficient Parallel Reads

Thu, 01 May 2025 00:00:00 GMT

Optimizing Partitioning for Efficient Parallel Reads Sragow, John Modern database management systems spend a significant portion of query execution time scanning data, so minimizing scanning latency is critical to maintaining high performance. As such, databases are partitioned into blocks so that queries can skip irrelevant tuples and avoid scanning the entire database. When this partitioning is optimized to minimize the number of blocks accessed by each query, smaller queries that access very few blocks fail to fully utilize the bandwidth because they cannot take advantage of parallel reading. However, reducing the size of each block in order to increase the number of blocks accessed by smaller queries slows down larger queries by forcing them to increase the number of I/Os they must perform. We propose a novel partitioning scheme that shuffles the row groups of blocks accessed by smaller queries so that they can read fewer tuples from multiple blocks in parallel without increasing the I/O cost of larger queries. Our experiments show that this technique allows smaller queries to be scanned up to twice as fast on larger block sizes as they would on a standard partitioning without significantly slowing down larger queries.

Integrating Functional Knowledge into Protein Design: A Novel Approach to Tokenization and Noise Injection for Function-Aware Protein Language Models

Thu, 01 May 2025 00:00:00 GMT

Integrating Functional Knowledge into Protein Design: A Novel Approach to Tokenization and Noise Injection for Function-Aware Protein Language Models Tang, Adrina Designing novel proteins with specific biological functions remains a fundamental challenge in computational biology. While recent advances in protein language models have enabled powerful sequence-based representations, most models, including state-of-the-art systems like ESM3, fall short in effectively encoding functional context during protein generation. In this work, we present a multimodal protein co-design framework that conditions sequence generation on fine-grained functional annotations, specifically leveraging residue-level Gene Ontology (GO) term labels on sequences from the UniRef100 database. By explicitly associating functional signals with residue elements of proteins, our model learns to generate function-conditioned protein sequences that are biologically plausible and semantically consistent. Unlike prior approaches, which treat function as a secondary feature or a classification task, our method focuses on joint reasoning over function and sequence during the design process. This closes a critical gap in the current landscape of protein design tools, offering a scalable and generalizable approach to co-designing protein sequences with user-specified functional profiles.

Pairwise Matching of Intermediate Representations for Fine-grained Explainability

Thu, 01 May 2025 00:00:00 GMT

Pairwise Matching of Intermediate Representations for Fine-grained Explainability Shrack, Lauren The differences between images belonging to fine-grained categories are often subtle and highly localized, and existing explainability techniques for deep learning models are often too diffuse to provide useful and interpretable explanations. We propose a new explainability method (PAIR-X) that leverages both intermediate model activations and backpropagated relevance scores to generate fine-grained, highly-localized pairwise visual explanations. We use animal and building re-identification (re-ID) as a primary case study of our method, and we demonstrate qualitatively improved results over a diverse set of explainability baselines on 35 public re-ID datasets. In interviews, animal re-ID experts were in unanimous agreement that PAIR-X was an improvement over existing baselines for deep model explainability, and suggested that its visualizations would be directly applicable to their work. We also propose a novel quantitative evaluation metric for our method, and demonstrate that PAIR-X visualizations appear more plausible for correct image matches than incorrect ones even when the model similarity score for the pairs is the same. By improving interpretability, PAIR-X enables humans to better distinguish correct and incorrect matches.

Enabling End-to-End Sensitivity Analysis of Integrated Models

Thu, 01 May 2025 00:00:00 GMT

Enabling End-to-End Sensitivity Analysis of Integrated Models Davidson, Rosemary K. As space-based precision-pointed telescopes continue to grow in scale and complexity, integrated models are increasingly relied upon to inform early design decisions and support system-level verification. When ground testing of full-system configurations is infeasible, integrated models, including structural-thermal-optical performance models, are essential for predicting performance and validating requirements across multidisciplinary, coupled domains. In early design phases, when uncertainty is high and design decisions have long-term implications for cost and schedule, it is especially important to understand which uncertain parameters most influence system performance. Global sensitivity analysis can help identify dominant uncertainty sources and inform decisions about model reduction, testing priorities, and resource allocation. However, the computational cost of applying global sensitivity analysis to integrated models often exceeds available resources. The presence of cross-disciplinary coupling between subsystem models further complicates analysis efforts. Coupled and dependent variables obscure how specific inputs influence system-level performance, limiting the ability to reduce model dimensionality or focus testing efforts on individual subsystems. There is a need for integrated modeling methodologies that enable tractable global sensitivity analysis of large, feedforward-coupled systems while preserving the accuracy needed to support early-phase design. This thesis develops both exact and approximate methods for performing global sensitivity analysis on integrated models. A set of exact propagation techniques is introduced to compute end-to-end sensitivity indices when specific structural conditions are met, including functional linearity, non-interacting transforms, and monotonic intermediate mappings. These methods are evaluated using a suite of benchmark test cases that isolate when the exact sensitivity analysis method is valid and when structural assumptions begin to break down. A modular modeling framework is developed to compute exact or approximate end-to-end sensitivity indices and to enable automated mapping between disciplinary models in the integrated chain. The approach is also applied to a representative linearized structural-thermal-optical performance model, demonstrating how end-to-end global sensitivity analysis can be performed efficiently across thermal, structural, and optical subsystems. To extend tractable sensitivity analysis to black-box models, several approximate strategies are introduced, including multifidelity surrogate modeling and statistical regression. These methods support both forward uncertainty propagation and variance-based global sensitivity analysis for structurally complex integrated models, without requiring full-system evaluation at every iteration. Together, the exact and approximate strategies developed in this work provide a foundation for scalable end-to-end global sensitivity analysis in early-phase design, where identifying influential parameters and constraining model complexity are essential for evaluating candidate architectures and informing mission decisions.

Deep Learning for Space Object Density Distribution Prediction

Thu, 01 May 2025 00:00:00 GMT

Deep Learning for Space Object Density Distribution Prediction Sarangerel, Sumiyajav The rapid growth of artificial objects in Low Earth Orbit (LEO) has heightened concerns over orbital congestion and collision cascades, known as Kessler Syndrome. Traditional high-fidelity models, while accurate, are computationally intensive and poorly scalable. This thesis introduces a machine learning–based framework for forecasting the long-term evolution of space object density. A large dataset is generated, using the MIT Orbital Capacity Assessment Tool – Monte Carlo (MOCAT-MC), simulating thousands of scenarios across varying launch, disposal, and maneuver parameters. A Convolutional Gated Recurrent Unit (ConvGRU) is trained to predict density distributions over a 100-year horizon, achieving accurate forecasts with significantly reduced runtime. With a simple guidance mechanism, the generalization capability of the model across diverse scenarios is greatly improved. This approach offers a scalable and efficient tool for supporting future space traffic management and sustainability efforts.

Towards Understanding Privacy Leakage in Decentralized and Collaborative Learning

Thu, 01 May 2025 00:00:00 GMT

Towards Understanding Privacy Leakage in Decentralized and Collaborative Learning Shi, Yichuan The emergence of large-scale machine learning (ML) models has highlighted a fundamental conflict: While computational demands push for the consolidation of data and models in vast, centralized data centers, real-world data continues to be distributed and fragmented across personal devices and private databases. How can we reconcile this contradiction without further monopolizing the ML ecosystem? What unique privacy and security risks arise from alternative ML orchestration system designs? Furthermore, how do these vulnerabilities and system failures inform our understanding of both how and what machines learn? This thesis attempts to explore these questions. It first examines key types of privacy leakages, evaluating their impact under realistic, cross-distribution settings. It then introduces a benchmarking analysis platform, SONAR, to investigate the relationship between privacy leakage (measured by attack performance), network topology, and data distribution. Finally, it presents Co-Dream, a novel algorithm for collaborative learning that offers improved privacy characteristics.

Prototyping a Scalable Proof Engine

Thu, 01 May 2025 00:00:00 GMT

Prototyping a Scalable Proof Engine Rosario, Jon Formal verification is an exciting development in software engineering, enabling implementations of programs to be rigorously checked against mathematical specifications. Assuming the specification is well-defined, formal verification provides guarantees of a program’s correctness and freedom from bugs that are simply not possible with test-based methods. There’s just one catch: the process of verifying large programs in popular theorem provers such as Coq (now known as Rocq) or Lean is painfully slow. These proof assistants rely on proof engines to construct proofs of correctness for given properties, but to our knowledge, there is no widely available proof engine that offers strong performance guarantees. Even more frustrating is the lack of consensus on what “good” performance should even mean in this context. This thesis lays the groundwork for addressing that gap by presenting a proof engine design that achieves asymptotically linear-time performance with respect to several important variables. We illustrate the design and its performance characteristics with examples from an implementation of the design and outline directions for future work.

Embedded Computing for Wavefront Control on Future Space Telescopes

Thu, 01 May 2025 00:00:00 GMT

Embedded Computing for Wavefront Control on Future Space Telescopes Belsten, Nicholas Future space telescopes will use adaptive optics to suppress starlight to directly image and characterize exoplanets. A measurement using this technique may be the first to detect extraterrestrial life in the universe. However, the real-time execution of adaptive optics control algorithms places unprecedented demands on spaceborne processors. Previous work has determined that processing limitations can degrade the achievable contrast and scientific yield of future exoplanet imaging missions. In this work, we quantify the relationship between adaptive optics processing needs and high contrast performance for the Habitable Worlds Observatory (HWO), a mission expected to launch in the 2040s and achieve the 10^-10 contrast necessary to image Earth-like planets around Sun-like stars. We survey the current landscape of high-order wavefront sensing and control (HOWFSC) algorithms for a future mission like HWO. We parameterize the compute requirements of multiple algorithms through analyses of computational patterns, benchmarks, and problem scaling. In parallel, we assess the capabilities of current and emerging spaceborne processors. We integrate these findings to model processor requirements across several dimensions of telescope design, and we predict whether various processor choices can meet the computational demands of specific HWO configurations. To validate our models, we implement HOWFSC algorithms on representative embedded processors and compare measured performance to predictions. These implementations also reduce risk for spaceflight by increasing the technology readiness level (TRL) of the algorithm–processor pairing to TRL 4. Given the significant uncertainty in HWO’s eventual design, we extend our deterministic models using Monte Carlo methods to evaluate system performance under uncertainty. We identify key sources of uncertainty and estimate the achievable contrast across a range of system configurations. Our results show that offloading computation to the ground is an important architectural option for most HWO designs. Even under optimistic assumptions, current space processors are insufficient to support the full range of HWO configurations. However, newly developed efficient algorithms substantially reduce the computational burden. Overall, we estimate that current technology has only a 40% probability of supporting HWO’s mission goals without additional architectural innovations. We conclude by recommending combinations of onboard computing, ground offloading, and optical design constraints to help close this technology gap as the mission design matures. In particular, we find that telescope stability and ground-in-the-loop performance are primary drivers of contrast performance, while algorithmic advances such as AD-EFC and onboard compute approaching ground-based GPU performance also provide significant benefits.

Stress-Guided Material Segmentation for Recycled 3D Printed Structures Using Finite Element Analysis

Thu, 01 May 2025 00:00:00 GMT

Stress-Guided Material Segmentation for Recycled 3D Printed Structures Using Finite Element Analysis Paulin, Cole J. We present a simulation-driven method for optimizing the structural performance of 3D printed objects made with recycled and fresh filament. Although sustainable materials such as recycled PLA reduce environmental impact, they often exhibit degraded or inconsistent mechanical properties, making them less suitable for structurally demanding applications. To address this, we develop a finite element analysis (FEA) pipeline that simulates stress and strain distributions under user-defined loading conditions, enabling intelligent segmentation of the object into regions of high and low mechanical demand. These segmented regions can be assigned recycled or fresh material during fabrication. Our system leverages open-source tools (SfePy) for simulation and we validate its accuracy against Abaqus, a commercial industry standard. We also introduce methods for automatically identifying and correcting segmentation artifacts, such as small disconnected islands. Through comparative simulation studies and performance evaluation, we demonstrate that our approach enables more sustainable 3D printing without sacrificing structural reliability

Metaheuristic Optimization for Automatic Arrangement of Power Electronics Components in a Shipboard Electrical Distribution System

Thu, 01 May 2025 00:00:00 GMT

Metaheuristic Optimization for Automatic Arrangement of Power Electronics Components in a Shipboard Electrical Distribution System Lohier, Sebastien This thesis proposes a novel methodology for the automatic placement of Power Electronics Building Blocks (PEBBs) in modular, integrated power corridor designs. These building blocks, which are created and tested offsite for a variety of applications, are currently placed manually during the design process, a method that is time-consuming and suboptimal. To address this challenge, we reduce the placement problem to a 2D bin-packing problem, leveraging a hybrid approach combining Genetic Algorithms and Simulated Annealing. This approach enables the generation of optimized placements that find the extremes of arbitrary heuristics, including minimizing routing distance and power density, effectively improving both design efficiency and system performance. The proposed methodology offers a significant step toward automating and optimizing the layout of power electronic components in complex systems.

An Interpretable Multimodal Framework for Regional Organ Transplantation Outcomes

Thu, 01 May 2025 00:00:00 GMT

An Interpretable Multimodal Framework for Regional Organ Transplantation Outcomes Lee, Ju Young The demand for kidney transplants continues to outpace supply, with over 89,792 patients on the waitlist as of September 2024, yet only 27,332 transplants performed in 2023 [1], and 28% of recovered kidneys going non-utilized [2]. In this thesis, we highlight the use of large language model (LLM) embeddings combined with structured tabular data to build a predictive classifier that estimates offer outcomes for kidney donor-recipient matches. For each predictive model deployed, we provide further analysis on the interpretability of these black-box models using a custom-designed SHAP analysis framework. Our study focuses on three distinct U.S. regions (Regions 1,2,3) with markedly different demographics and amounts of data on organ acceptances (Region 1: 43,126 offers with 2.19% acceptance rate, Region 2: 394,640 offers with 1.57% acceptance rate, Region 3: 169,342 with 2.23% acceptance rate in years 2016-2019). Among the baseline XGBoost models, Region 3 achieved the highest performance, with a precision-accept score of 0.929 and accuracy of 0.993 in the test data. Building on this strong foundation, the multimodal TabText model in Region 3 achieved the best performance overall, with a precision-accept score of 0.959 and accuracy of 0.993 after fine-tuning for six epochs. Our findings suggest that increasing the number of text features, extending training epochs, and incorporating explicit numerical values led to improved model performance in Region 3. In Regions 1 and 2, the baseline model outperformed the TabText model, suggesting that data sparsity in these regions may have limited the effectiveness of the multimodal approach and that further hyperparameter tuning is needed. We also present several visualization techniques to enhance model interpretability. Specifically, we developed a novel SHAP explainer that illustrates feature interactions between multimodal inputs, including both tabular and textual data. Additionally, we explored methods to identify regions of high and low model fidelity by mapping per-sample prediction errors onto t-SNE embeddings. Overall, this thesis introduces new directions for transplant research in the context of transformer-based models and interpretable AI. Leveraging data-driven decision-support tools and refining allocation policies are essential steps toward addressing the persistent gap between supply and demand in the kidney transplant landscape.

Medium Access Control Protocol for Satellite Constellations

Thu, 01 May 2025 00:00:00 GMT

Medium Access Control Protocol for Satellite Constellations Li, Brian Satellite internet constellations have emerged as a promising solution for providing global internet connectivity, especially in regions underserved by terrestrial infrastructure. However, as user demand increases, especially in densely populated urban areas, existing Medium Access Control (MAC) protocols face significant scalability challenges and fail to take advantage of advanced antenna processing techniques, including phased array nulling, as well as capacity sharing via inter-satellite links. We present both an offline linear program and a novel online greedy MAC protocol to assign satellite resources to users using either sequential service, capacity sharing, or interference-aware nulling. Our offline formulation provides an upper bound on system performance, and while our online protocol is sub-optimal compared to this optimum, it is designed to be implementable on a real-time system. Simulations demonstrate that incorporating nulling can increase effective capacity by up to 25 times, substantially boosting profit in high-demand scenarios. We further quantify the performance gap between the online protocol and the offline optimum under varying demand distributions, showing that our online approach achieves near-optimal results in low-peakiness settings and gracefully degrades under more extreme conditions. These results highlight the importance of spatial processing at the MAC layer and offer practical design insights for future satellite internet constellations.

Copilot Tutor: Automated Software Engineering Practice Augmented with LLMs

Thu, 01 May 2025 00:00:00 GMT

Copilot Tutor: Automated Software Engineering Practice Augmented with LLMs Kong, Blisse In recent years, large language models (LLMs) have become more ubiquitous in the workplace. In software engineering, they are often realized as “copilots" which produce code given a prompt or existing code. Programmers using these tools to increase their coding productivity need to be proficient in inspecting and in understanding these copilots’ outputs. As engineers incorporate these tools to accelerate their workflows, they have a parallel opportunity to accelerate learning new programming languages. This thesis presents a tutor interface where students with some programming experience in an origin language can learn a target language while practicing how to critically read and fix a copilot’s output to write correct, safe programs. This work also introduces the automatic generation of exercises teaching syntax and semantics on which a programmer experienced in the origin language but not the target language should focus.

Strategic Physical Withholding of Renewable Energy Generators

Thu, 01 May 2025 00:00:00 GMT

Strategic Physical Withholding of Renewable Energy Generators Irvine, Paul M. Renewable generators may have incentives to strategically withhold energy output in electricity markets, either to exercise market power or to avoid congestion pricing caused by transmission constraints. Although academic work often treats renewables as not downward dispatchable, renewable generators technically can, at least in principle, reduce their output by self-curtailing. This paper shows that a firm with a large, diverse portfolio could find it profit-maximizing to withhold renewables over conventional thermal generators once it accounts for constraints on ramp rates and minimum generation, as well as the costs associated with starting-up generators and the probability of detection on generator type by market monitoring authorities. Long-term forward contracts like pay-as-produced Power Purchase Agreements (PPAs) can blunt incentives to exercise market power by insulating individual generators from wholesale prices; however, since generators under PPAs typically bid into the wholesale market and influence competitive prices, they may actually encourage renewable withholding if contract prices are sufficiently low and the parent firm’s portfolio is exposed to wholesale prices. To screen for renewable withholding, this paper proposes three methods: (1) examining the distribution of aggregate output across export interfaces for suspicious bunching, (2) testing deviations from ex-ante forecasts, and (3) identifying the time intervals where generators encounter model structural changes compared to a benchmark presumed free of withholding. Together, this work prepares academics and regulators to more accurately model the behavior of renewable generators in electricity markets and to screen for potential market abuses.

Argos: Verifiable FHE Using Commodity Hardware

Thu, 01 May 2025 00:00:00 GMT

Argos: Verifiable FHE Using Commodity Hardware Jepsen, Fisher We present Argos, a simple approach for adding verifiability to fully homomorphic encryption (FHE) schemes using trusted hardware. Traditional approaches to verifiable FHE require expensive cryptographic proofs, which incur an overhead of up to seven orders of magnitude on top of FHE, making them impractical. With Argos, we show that trusted hardware can be securely used to provide verifiability for FHE computations, with minimal overhead relative to the baseline FHE computation. An important contribution of Argos is showing that the major security pitfall associated with trusted hardware, microarchitectural side channels, can be completely mitigated by excluding any secrets from the CPU and the memory hierarchy. This is made possible by focusing on building a platform that only enforces program and data integrity and not confidentiality (which is sufficient for verifiable FHE, since all data remain encrypted at all times). All secrets related to the attestation mechanism are kept in a separate coprocessor (e.g., a TPM)—inaccessible to any software-based attacker. Relying on a discrete TPM typically incurs significant performance overhead, which is why (insecure) software-based TPMs are used in practice. As a second contribution, we show that for FHE applications, the attestation protocol can be adapted to only incur a fixed cost. Argos requires no dedicated hardware extensions and is supported on commodity processors from 2008 onward. Our prototype implementation introduces 3% overhead for FHE evaluation, and 8% for more complex protocols. In particular, we show that Argos can be used for real-world applications of FHE, such as private information retrieval (PIR) and private set intersection (PSI), where providing verifiability is imperative. By demonstrating how to combine cryptography with trusted hardware, Argos paves the way for widespread deployment of FHE-based protocols beyond the semi-honest setting, without the overhead of cryptographic proofs.

Automatic Conversion of C and C++ Programs to the BuildIt Multi-Stage Programming Framework

Thu, 01 May 2025 00:00:00 GMT

Automatic Conversion of C and C++ Programs to the BuildIt Multi-Stage Programming Framework Kumar, Aryan BuildIt allows users to write C++ programs that can execute in multiple stages, where the output of one stage is the program source for the next stage, ending with some final output produced. This is particularly useful for writing specialized code and generating code for domain-specific languages. While there are other approaches to multi-stage programming, BuildIt has several advantages: it takes a library-based approach (so it requires no modifications to the compiler and is thus highly portable), and it has excellent ease of use as all the user has to do is change the declared types of variables in their C++ program. The goal of this thesis is to further improve BuildIt’s ease of use by simplifying this step: in particular, by developing a tool that will automatically convert existing C and C++ programs to the BuildIt framework. We show how to use Clang tooling in conjunction with modifications to the Clang compiler to perform non-trivial modifications to source, namely type-modification, to automatically convert code to its unstaged BuildIt equivalent. As the unstaged BuildIt code can be specialized by staging certain variables, this tool will ultimately enable more easily staging and optimizing C/C++ repositories with the BuildIt framework.

Association of GLP-1 Receptor Agonist Use with Kidney and Cardiovascular Outcomes in Stable Kidney Transplant Recipients

Thu, 01 May 2025 00:00:00 GMT

Association of GLP-1 Receptor Agonist Use with Kidney and Cardiovascular Outcomes in Stable Kidney Transplant Recipients Jung, Emma Yejoo Recent surges in the use of glucagon-like peptide-1 receptor agonists (GLP-1RA) have shown promise in reducing cardiovascular events and improving kidney function in patients with type 2 diabetes. Due to these hopeful improvements, kidney transplant recipients (KTRs) have started using GLP-1RA. However, their effects in KTRs remain largely unstudied in clinical studies. This thesis uses a large-scale Electronic Health Record (EHR) database to perform a retrospective cohort analysis to study the association between GLP-1RA use and kidney and cardiovascular outcomes amongst stable KTRs. Primary outcomes include all-cause mortality, major adverse kidney events (MAKE), and major adverse cardiac events (MACE). Among stable KTRs, GLP-1RA users show reduced risk for all-cause mortality (adjusted hazard ratio [aHR]: 0.45; 95% confidence interval [CI]: 0.32-0.62) and MAKE (aHR: 0.69; 95% CI: 0.58-0.81), but no significant difference for MACE (aHR: 0.84; 95% CI: 0.67-1.05). In addition, users show increased risk for irritable bowel syndrome (IBS) (aHR: 2.11; 95% CI: 1.07-4.15) and urinary tract infection (UTI) (aHR: 1.53; 95% CI: 1.27-1.85). These results indicate the potential of GLP-1RA to reduce mortality and adverse kidney outcomes and increase IBS and UTI in KTRs.

Hardware Acceleration for Real-Time Compression of 3D Gaussians

Thu, 01 May 2025 00:00:00 GMT

Hardware Acceleration for Real-Time Compression of 3D Gaussians Kahler, Kailas B. 3D Gaussian Splatting (3DGS) is a technique for novel view synthesis, where images of a scene from a specific viewpoint are generated using images from different viewpoints, that has gained popularity for its reduced computational overhead, resulting in faster training and rendering times compared to other methods like Neural Radiance Fields (NeRFs). Its applications outside of strictly novel view synthesis have also been explored, with monocular simultaneous localization and mapping (SLAM) in robotics being an emergent application. However, because of limited on-board battery capacity, the computer hardware used in small robots is much less capable than the high-powered GPUs that the 3DGS algorithm was originally developed on, having both less compute and memory capacity and bandwidth. While there has been work developing specialized compute for the rendering pipeline of 3DGS, memory remains an obstacle to deployment. The Gaussian map can occupy from 1MB − 700MB in memory, which is both too large to store on-chip within micro-robots and such that moving Gaussians from memory to compute can dominate power consumption. While there has been prior work on algorithms for compressing Gaussian representations, they are not yet capable of running in real-time on the hardware present in these robots, as would be required for SLAM. Thus, this thesis explores the limits of these compression methods on current hardware, resulting in an optimized CUDA implementation with better than 100× the throughput of prior work and achieving real-time operation on workstation-class hardware. However, after concluding that custom hardware is necessary for further improvement, this thesis also presents a hardware accelerator that nears real-time compression performance within a reduced power budget, outperforming an NVIDIA Jetson Orin Nano with 64% higher throughput while using 1/16th of the multipliers and drawing 38% of the power when running at 100MHz on an AMD UltraScale+ FPGA.

Personalization of AI Tutor Based on Knowledge Graphs

Thu, 01 May 2025 00:00:00 GMT

Personalization of AI Tutor Based on Knowledge Graphs Huang, Sheng Personalized tutoring, tailored to the specific knowledge and needs of individual students, has been shown to significantly enhance academic performance. Research by Schmidt and Moust, for example, highlights that tutors who engage with students on a personal level are more effective in guiding them toward higher academic achievement [1]. Inspired by this principle, the Axiom group at the MIT Media Lab developed an AI tutor for their Intro to Programming courses. The initial version of the tutor, RAGS, relied on analyzing past conversations between students and the tutor, as well as course content, to generate personalized responses. While this approach showed promise, it faced scalability challenges, such as the need to store an ever-growing volume of conversation history and the risk of exceeding token limits in prompt context windows. Additionally, the model occasionally struggled with over-generalization, particularly when responding to vague questions based solely on historical interactions. To address these limitations, this thesis introduces a new approach: a student knowledge graph. Rather than relying on an expanding archive of past conversations, the knowledge graph uses weighted nodes to represent a student’s understanding of each concept. A weight of -8 indicates subpar understanding, while a weight of 8 signifies mastery. After pre-processing the course data, the graph maintains a fixed size, eliminating the need for additional storage over time. This innovation solves two critical problems: 1. Scalability: By leveraging a fixed-size PostgreSQL database, the student knowledge graph avoids the storage challenges associated with saving endless conversation histories. 2. Improved Personalization: Instead of sifting through old conversations, the tutor uses concept weights to generate more precise and contextually relevant responses, even to vague questions. Testing and evaluation of the implemented system demonstrate its effectiveness in both scalability and response quality. Over 60% of survey participants reported that the knowledge graph-enhanced tutor provided clearer and more relevant guidance, particularly when building on concepts they already understood. Additionally, over 80% of respondents noted improvements in the tutor’s ability to address weak areas and provide targeted practice, especially when preparing for quizzes or exams.

SPIRAL: Iterative Subgraph Expansion for Knowledge-Graph Based Retrieval-Augmented Generation

Thu, 01 May 2025 00:00:00 GMT

SPIRAL: Iterative Subgraph Expansion for Knowledge-Graph Based Retrieval-Augmented Generation Hadjiivanov, Michael D. Large language models (LLMs) excel at generating fluent answers but are prone to hallucination when the prompt fails to anchor them to verifiable facts. Retrieval-augmented generation (RAG) mitigates this risk, yet existing graph-based retrievers either return bloated neighborhoods or incur prohibitive latency on large knowledge graphs (KGs). We introduce SPIRAL—Supervised Prior + Iterative Reinforcement with Adaptive Labelling—a lightweight two-stage framework that constructs compact, tree-shaped evidence subgraphs. This differs from previous work in its use of a trained, iterative policy network built on top of a prior over triples, delivering improved performance on multi-hop question answering tasks. Stage 1 trains a single-label GLASS-GNN on shortest-path heuristics, producing frozen, question-aware node embeddings at negligible runtime cost with significant local topology awareness around question entities. Stage 2 layers a GLASS policy—which re-labels the partial subgraph at each step—on top of these embeddings and optimizes it with proximal policy optimization. The policy scores only the 1-hop frontier, enabling sub-second inference even on million-edge graphs. On the multi-hop KGQA benchmark WebQSP, SPIRAL attains 0.95 triple recall and 0.97 answer recall while retrieving at most 50 triples—doubling the sampling efficiency of the strongest prior work. Coupled with Llama 3.1-8B, the retrieved trees boost Hit@1 by 2.5 % over SubgraphRAG. Ablation studies confirm that adaptive labels are critical for multi-hop reasoning. SPIRAL demonstrates that accurate and concise retrieval is achievable without resorting to massive models or expensive graph crawls, opening the door to real-time, KG-grounded assistants on modest hardware.

Injection of Domain-Specific Knowledge for Enterprise Text-to-SQL

Thu, 01 May 2025 00:00:00 GMT

Injection of Domain-Specific Knowledge for Enterprise Text-to-SQL Choi, Justin J. This work examines the current state of using large language models (LLMs) to solve Text-to-SQL tasks on databases in an enterprise setting. Benchmarks on publicly available datasets do not fully capture the difficulty and complexity of this task in a real-world, enterprise setting. This study examines the critical steps needed to work with enterprise data as well as using knowledge-injection to enhance the performance of LLMs on Text-to-SQL tasks. We begin by evaluating the baseline performance of LLMs on enterprise databases, revealing that a predominant source of failure stems from a lack of domain-specific knowledge. To improve performance, we explore knowledge-injection: the process of incorporating internal and external knowledge. Internal knowledge consists of database-specific information such as join logic, while external knowledge refers to institutional acronyms or group names. We present a hybrid retrieval pipeline that combines embedding and text based searching with LLM-guided ranking to supply models with relevant external knowledge during Text-to-SQL generation. We evaluate the impact of the knowledge-injection by testing the performance of LLMs on the table retrieval task after being augmented with appropriate external knowledge. We demonstrate that knowledge-injection significantly improves accuracy on table retrieval using BEAVER: an enterprise-level Text-to-SQL benchmark. Our findings highlight the importance of domain-specific knowledge-injection and retrieval augmentation in bringing LLMs closer to deployment in enterprise-grade database systems, as well as common failure modes that occur when executing enterprise Text-to-SQL.

Evaluating the Feasibility of Transaction Scheduling via Hardware Accelerators

Thu, 01 May 2025 00:00:00 GMT

Evaluating the Feasibility of Transaction Scheduling via Hardware Accelerators Chomphoochan, Thanadol As single-thread performance plateaus, modern systems increasingly rely on parallelism to scale throughput. Yet, efciently managing concurrency—particularly in transactional systems—remains a major bottleneck. This thesis explores the feasibility of accelerating transaction scheduling via hardware, leveraging FPGAs to ofoad scheduling logic from the CPU. We revisit Puppetmaster, a hardware transaction scheduler, and present a redesigned architecture emphasizing deployability, modularity, and evaluation. We implement both an optimized software baseline and a Bluespec-based hardware design, evaluating their performance across synthetic YCSB-style workloads with varying contention levels. Our hardware prototype demonstrates competitive throughput, achieving over 90% of peak throughput even under high-contention workloads. These results validate the potential of transaction scheduling as a target for hardware acceleration and highlight promising directions for future hybrid hardware-software concurrency-control systems.

Computational Exploration of Thermodynamic Models of Geological CO₂ Injection

Thu, 01 May 2025 00:00:00 GMT

Computational Exploration of Thermodynamic Models of Geological CO₂ Injection Edelman, Jonathan This thesis investigates the behavior of carbon dioxide flow in porous media through high-fidelity computational modeling, with a specific focus on the impact of the Span-Wagner equation of state (EOS). Accurate modeling of CO₂ transport in subsurface environments is essential for applications such as carbon capture and storage (CCS). We model the entire flow from injection, down throughout a vertical pipe and into a porous reservoir. To this end, we utilize the MOOSE (Multiphysics Object-Oriented Simulation Environment) framework developed by Idaho National Laboratory to perform finite element simulations. A key contribution of this work is the successful coupling of a porous rock domain with a one-dimensional pipe flow simulation in Julia, enabling a broader representation of injection scenarios. The study examines how the thermodynamic accuracy of the Span-Wagner Equation of State influences flow characteristics, in comparison to the Ideal Gas Equation of State. Through a series of coupled pipe-reservoir simulations, we assess variations in pressure and density as CO₂ is injected from the pipe into the porous medium. The model can detect phase change conditions, allowing us to predict the maximum mass flux that can be achieved below the liquefaction threshold, as defined by the binodal curve in the CO₂ phase diagram at a given temperature. The results highlight the importance of EOS selection in predicting multiphase flow behavior, especially under conditions relevant to geological storage. Furthermore, we find that the Ideal Gas EOS underpredicts injection rates under the same conditions. This integrated modeling approach advances the understanding of thermodynamic effects in coupled subsurface flow systems and supports the development of reliable tools for large-scale carbon storage applications.

Clinical Text De-identification Using Large Language Models: Insights from Organ Procurement Data

Thu, 01 May 2025 00:00:00 GMT

Clinical Text De-identification Using Large Language Models: Insights from Organ Procurement Data Dahleh, Omar This thesis presents a novel approach to the de-identification of clinical notes from Organ Procurement Organization (OPO) records, leveraging advanced natural language processing (NLP) methodologies. Specifically, we employ in-context learning using large language models (LLMs) to effectively identify and remove protected health information (PHI), aiming to maintain high data utility post-redaction. Our work systematically evaluates the performance of the LLM-based method against established baseline techniques, including traditional Named Entity Recognition (NER) and rules-based systems. Through a slew of experiments, we assesses the strengths and limitations of each method regarding precision and recall. This work will contribute to a uniquely extensive dataset, comprising millions of de-identified OPO clinical notes, which will facilitate ethical healthcare research and enhance compliance with contemporary data protection standards. Ultimately, this dataset holds significant potential for improving processes and outcomes within the field of organ donation and procurement.

Accelerating Novel Energy Catalyst Discovery Using Automation, Active Learning, and AI

Wed, 01 May 2024 00:00:00 GMT

Accelerating Novel Energy Catalyst Discovery Using Automation, Active Learning, and AI Ren, Zhichu The discovery of novel energy catalysts is a critical challenge in the field of materials science. Traditional methods for materials discovery are labor-intensive and time-consuming, hindering the rapid development of new catalysts. To address this issue, we introduce a comprehensive approach that integrates automation, active learning, and artificial intelligence (AI) to accelerate the discovery process. Our approach introduces the Copilot for Real-world Experimental Scientist (CRESt) system, which combines a large multimodal model (LMM) with an active learning-guided robotic system. CRESt streamlines the workflow of composition selection, high-throughput materials synthesis, electrochemical screening and characterization for the optimization of high-entropy alloy catalysts. The system allows researchers, regardless of their programming skills, to interact with the robotic platform using voice commands, making it highly accessible and user-friendly. We demonstrate the effectiveness of our approach by experimentally exploring over 700 chemistries and 1300 samples. The optimized 8-dimensional alloy (Pd-Pt-Cu-Au-Ir-Ce-Nb-Cr) achieved approximately 10 times the cost-specific performance of commercial catalysts for the direct formate fuel cell. This breakthrough highlights the potential of our approach to accelerate the discovery of novel energy catalysts across various domains. Furthermore, we discuss the challenges and considerations associated with implementing active learning in real-world experiments. We provide guidance on addressing model-centric and data-centric issues, such as model customization and data irreproducibility, to ensure the successful application of active learning in materials research projects. Looking ahead, we explore the role of human experimentalists in the era of AI-driven discovery. While AI and automation are poised to transform many aspects of experimental research, we argue that human experimentalists remain irreplaceable for now. Our ability to exercise critical thinking and engage in complex real-world interactions sets us apart from abiotic intelligence. However, as AI becomes more deeply integrated into research practices, the experimental landscape is bound to undergo significant changes.

Efficient ML Inference via Matrix-Vector Approximations

Thu, 01 May 2025 00:00:00 GMT

Efficient ML Inference via Matrix-Vector Approximations Li, Daniel D. Efficient inference is a growing priority in deep learning, where large model sizes and increasing deployment demands pose challenges for latency, memory, and energy usage. This thesis presents a unified framework for evaluating approximation methods that accelerate inference by modifying weight matrices. We model each method as a function ƒ_c(A) that approximates a weight matrix A under a compression rate c, and assess its impact on both matrix–vector accuracy and downstream task performance. We conduct empirical evaluations across two representative models, AlexNet on CIFAR10 and DistilBERT on AG News, comparing quantization, sparsification, and low-rank approximations. Our analysis spans four perspectives: (1) how different methods trade off ℓ₂ error and compression, (2) how weight statistics and input distributions shape error, (3) how well ℓ₂ error predicts classification accuracy, and (4) how idealized compression differs from real memory savings. We find that sparsification offers a strong trade-off between storage and accuracy, particularly because it preserves task-relevant structure in the weights. We also show that ℓ₂ error is not always a reliable proxy for accuracy, especially when input data lie on low-dimensional manifolds. These results suggest that approximation quality must be evaluated not only by global distortion metrics, but also by how the method interacts with model structure and input distributions. Our findings offer practical guidance for deploying efficient deep learning models and shed light on how compression affects performance in real-world settings.

A Pedagogical Multimodal System for Mathematical Problem-Solving and Visual Reasoning

Thu, 01 May 2025 00:00:00 GMT

A Pedagogical Multimodal System for Mathematical Problem-Solving and Visual Reasoning Lee, Jimin Effective reasoning often requires more than text or language. It requires visualizing, drawing, gesturing, and interacting for both humans and artificial intelligence (AI). Specifically in educational subjects, such as geometry and graphs, visual tools like auxiliary annotations and drawings can greatly help students understand abstract theories. This thesis explores and suggests how multimodal interaction between humans and AI helps humans engage with the system more naturally and effectively, leading to improved problem-solving in mathematical settings. Recent large multimodal models (LMMs) have the ability to facilitate collaborative reasoning by supporting textual, visual, and interactive inputs, diversifying methods of communication between humans and AI. Utilizing such advancements, this thesis also dives into the development of Interactive Sketchpad, a tutoring system that combines language-based explanations with interactive visualizations to enhance learning. It also reviews findings from user studies with Interactive Sketchpad, demonstrating that multimodality contributes to user task comprehension and engagement levels. Together, these contributions can reframe the role of AI in education as a visual and interactive collaborator that supports deeper reasoning rather than simply providing answers. Furthermore, this work demonstrates the potential of multimodal human-AI systems in fostering engagement and scaling personalized, visual learning across domains.

Fast and Scalable Subgraph Learning

Thu, 01 May 2025 00:00:00 GMT

Fast and Scalable Subgraph Learning Liang, Derrick Graph Neural Networks (GNNs) are a powerful framework for learning over structured data, enabling predictive modeling across domains such as bioinformatics, recommendation systems, and financial fraud detection. While scalable systems like SALIENT++ have advanced the training of node-level GNN tasks at industrial scale, they do not support an emerging class of workloads: subgraph classification, which is increasingly common in real-world applications. Prior implementations address this gap by modifying both the data pipeline and the model architecture—but at the cost of composability, creating tightly coupled systems that slow further development. This thesis introduces MOSAIC, a lightweight data transformation that reframes subgraph classification as nodewise prediction by augmenting the graph with representative nodes. This approach enables direct compatibility with SALIENT++ and other nodewise systems while decoupling workload format, dataloader design, and model architecture. I demonstrate that MOSAIC enables modular reuse of architectures like GraphSAGE and subgraph-aware components from GLASS, while preserving SALIENT++’s system-level scalability. On the large-scale Elliptic2 dataset, this integration reduces training memory usage by 2.8× and epoch runtime from over 90 minutes to 0.4 seconds—while improving classification performance. I implement MOSAIC as a succinct (<100-line), reusable preprocessing script, enabling integration of the GLASS architecture into SALIENT++ in <10 lines of code, compared to Wang et al.’s tightly coupled 500+ line design. These results highlight the feasibility of scalable, composable experimentation for subgraph learning tasks in high-performance GNN systems.

Dynamic Scene Editing via Semantically Trained 3D Guassians

Thu, 01 May 2025 00:00:00 GMT

Dynamic Scene Editing via Semantically Trained 3D Guassians Lam, Jordan Image-based 3D scene reconstruction continues to be a challenge as it involves solving both the sufficient 3D representation problem and the 3D reconstruction itself. One approach to tackle the rendering problem is 3D Gaussian Splatting because of its potential to produce fast and realistic renders via 3D Gaussian representation. With many applications in the entertainment industry, there is motivation in using 3D Gaussian Splatting for not only reconstructing 3D dynamic scenes but also editing them. However, extending the problem to dynamic 3D scenes proves to be a challenging task as it involves discerning the correct representation of a 3D scene while maintaining the capability to render in real time. State-ofthe-art methods have proposed methods that reconstruct dynamic scenes or edit static scenes, but the problem of editing dynamic scenes is still underexplored. This thesis analyzes the feasibility of editing semantically trained Gaussians for dynamic 3D scene editing. By training 3D Gaussians to represent the semantics across the time steps of a dynamic 3D scene, these primitives can be combined with an image editing pipeline to perform real-time, realistic 3D scene editing. Results show that editing segmented 3D Gaussians produces higher-quality and efficient renders as compared to editing without segmentation. However, when evaluated for mainstream applications, results show the impracticality of this pipeline and draw focus to memory and editing limitations that need to be further researched for future advances in 3D Gaussian Splatting.

Decentralized AI for Methylation Data with Applications to Precision Health

Thu, 01 May 2025 00:00:00 GMT

Decentralized AI for Methylation Data with Applications to Precision Health Jamee, Mehrab S. Advances in precision health rely on integrating large-scale genomic data to identify biomarkers and predict health outcomes. However, sharing sensitive patient data between institutions like hospitals poses significant privacy and security challenges, limiting collaboration and the development of robust machine learning models. This thesis proposes a decentralized artificial intelligence framework for analyzing DNA methylation data, enabling institutions to collaboratively train models without exchanging sensitive information. By taking advantage of generative deep learning techniques and federated learning paradigms, the framework aims to impute missing biomarkers in fragmented datasets and improve the accuracy of downstream predictive tasks, like predicting chronological age, mortality, and cancer data. Two intermediate models are implemented and evaluated in this thesis. The first predicts age from DNA methylation data, and can be used for evaluation of the imputation model. The second is an imputation model that uses a conditional autoencoder architecture to reconstruct missing biomarker data in clinical datasets, which is designed to take advantage of contextual methylation embeddings, made available by recently published pretrained epigenomics foundation models. This work seeks to advance the use of decentralized AI in epigenomics, with the ultimate goal of improving personalized healthcare while preserving patient privacy.

Exploring Smallholder Field Delineation

Thu, 01 May 2025 00:00:00 GMT

Exploring Smallholder Field Delineation Janjigian, Lily T. Accurate crop field delineation from satellite imagery is a critical component of agricultural monitoring. However, most existing models are developed and evaluated in large-scale, industrial agricultural regions, where field boundaries are relatively regular and high-quality annotated data is more readily available. In contrast, smallholder regions—where fields are smaller, more irregularly shaped, and often lack precise geospatial labels—remain underrepresented in both data and model performance. This thesis investigates model architectures, loss functions, and learning paradigms for improving segmentation performance in smallholder settings. Using datasets from Austria, India, and Rwanda, we evaluate several model configurations including ResUNet++ with Dice+BCE and Tanimoto+BCE losses, a meta-learned ResUNet++ using Model-Agnostic Meta-Learning (MAML), and SAM2 ViT-H, a large vision transformer released by Meta, evaluated in a zero-shot setting. We introduce a data processing pipeline that converts vector field boundaries from the FTW dataset into highresolution image–mask pairs suitable for supervised learning. Quantitative and qualitative results reveal that models trained on industrial-scale data perform poorly in smallholder regions without adaptation. SAM2 exhibits strong zero-shot performance, especially on larger fields, while ResUNet++ models trained directly on India perform more consistently across small-to-medium sized fields. MAML yielded underwhelming performance under resource constraints, highlighting the need for further tuning. These findings underscore the importance of geographically diverse, well-aligned training data and support the case for developing globally representative agricultural segmentation datasets.

You Only Look Twice: An Ensemble Deep Learning Model for Wildfire Detection Using Terrestrial Camera Networks

Thu, 01 May 2025 00:00:00 GMT

You Only Look Twice: An Ensemble Deep Learning Model for Wildfire Detection Using Terrestrial Camera Networks Jones, John M. Wildfires represent a growing global threat that requires rapid detection and response to minimize environmental damage, economic losses, and human casualties. In the United States, California stands out as a particularly common wildfire hot spot. Recent fire seasons have shattered historical records and been particularly devastating. This work investigates innovative methods for classifying and localizing wildfires through terrestrial cameras positioned on elevated terrain, aimed at improving early detection capabilities and response times while maintaining computational efficiency and reliability for the U.S. Space Force in Southern California. We present YOL2, a novel ensemble approach that combines a fine-tuned ConvNeXt Convolutional Neural Network incorporating a Dynamic Tanh normalization layer with a fine-tuned YOLO11 model for precise localization. Using a comprehensive dataset of 33,636 time-sequenced images from terrestrial cameras across the United States and Europe, our system achieves 98% fire detection accuracy and 55% localization mean average precision [50:95]. The implementation of Dynamic Tanh normalization—applied for the first time in wildfire detection—enhances computational efficiency without sacrificing performance. The images used capture the spread of incipient fires over time, with most containing bounding boxes denoting the approximate location of fire, allowing our system to identify fires quickly while minimizing false positives. Importantly, our spatiotemporal system operates effectively without requiring individual models to rely on multiple time steps as input, enabling modular component replacement and adaptation. The use of pan, tilt, and zoom cameras in concert with our YOLO model provides a more computationally efficient confirmation of fire than alternative methods, showing that extracting better results from less information is possible. Beyond wildfire applications, the YOL2 ensemble methodology demonstrates profound implications for remote sensing more broadly. This work establishes a foundation for highly efficient visual detection systems applicable across numerous domains requiring rapid and accurate object identification and localization.

Towards transparent representations: on internal structure and external world modeling in LLMs

Thu, 01 May 2025 00:00:00 GMT

Towards transparent representations: on internal structure and external world modeling in LLMs Hariharan, Kaivalya Large language models (LLMs) generalize far beyond their training distribution, enabling impressive downstream performance in domains vastly different from their pretraining distribution. In this thesis, we develop a data-centric view on machine learning. We suggest that the deep generalization of LLMs is best understood through studying the relationships between the four fundamental components of this data generalization: pretraining data, test-time inputs, model outputs, and internal structure. Of these, we present two full research studies characterizing test-time inputs and internal structure. Chapter 1 develops the data-centric view of machine learning, and outline the thesis. Chapter 2 presents Breakpoint, a method of generating difficult coding tasks for models at a large scale that attempts to disambiguate the factors that make problems at test-time difficult. Chapter 3 analyzes the structure of gradient-based jailbreaks in LLMs. We argue that even though GBJs are more out of distribution than even random text, they induce a low-rank, structured change in models. Finally, Chapter 4 discusses the recent rise of reasoning models and proposing some lines of future work in the data-centric view towards developing more robust understanding of LLMs.

Biomechanical Validation of Skeletal Tracking Data and Developing Action Recognition Models for Basketball: A Baseline for NBA Officiating Tools

Thu, 01 May 2025 00:00:00 GMT

Biomechanical Validation of Skeletal Tracking Data and Developing Action Recognition Models for Basketball: A Baseline for NBA Officiating Tools Hong, Stephen S. Optical tracking technology in sports has advanced rapidly in recent years, enabling new opportunities for data-driven analysis and tools to enhance the game. This study presents a framework for processing and analyzing a new skeletal tracking dataset collected from NBA basketball games. The methodology includes biomechanical joint validation, anomaly detection, and region-based consistency analysis to assess the integrity of player motion data. Joint movement anomalies are used to detect tracking errors, while court region and stadium-level evaluations help identify where the optical tracking system may be underperforming. These patterns can guide data providers toward specific areas that require refinement, offering a clearer starting point for improving system accuracy. After cleaning the dataset of 117 NBA games, two action recognition models—a transformer-based model and a temporal graph neural network—are implemented to classify player actions, specifically dribbling, passing, shooting, and rebounding, from sequences of skeletal tracking frames. The objective is to establish a baseline for developing tools to support officiating decisions in the NBA. By leveraging spatiotemporal representations of joint motion, this work improves the reliability of skeletal tracking data and contributes to the advancement of automated decision support in professional sports officiating.

Towards Enhanced Proposals for PINN-Based Neural Sampler Training

Thu, 01 May 2025 00:00:00 GMT

Towards Enhanced Proposals for PINN-Based Neural Sampler Training Erives, Ezra Sampling from distributions whose density is known up to a normalizing constant is an important problem with a wide range of applications including Bayesian posterior inference, statistical physics, and structural biology. Annealing-based neural samplers seek to amortize sampling from unnormalized distributions by training neural networks to transport a family of densities interpolating from source to target. A crucial design choice in the training phase of such samplers is the proposal distribution by which locations are generated at which to evaluate the loss. Previous work has obtained such a proposal distribution by combining a partially learned vector field with annealed Langevin dynamics. However, isolated modes and other pathological properties of the annealing path imply that such proposals achieve insufficient exploration and thereby lower performance post training. In this work we extend existing work and characterize new families of proposals based on controlled Langevin dynamics. In particular, we propose continuously tempered diffusion samplers, which leverage exploration techniques developed in the context of molecular dynamics to improve proposal distributions. Specifically, a family of distributions across different temperatures is introduced to lower energy barriers at higher temperatures and drive exploration at the lower temperature of interest. We additionally explore proposals based on Langevin dynamics involving non-Newtonian kinetic energies. We empirically validate improved sampler performance driven by extended exploration.

From Sketch to Stage: Tools for Prototyping and Exporting Collaborative DMIs on the Web

Thu, 01 May 2025 00:00:00 GMT

From Sketch to Stage: Tools for Prototyping and Exporting Collaborative DMIs on the Web Luchko, Yaro This thesis presents tools and ideas for prototyping and exporting collaborative digital music instruments (DMIs) on the web, the primary purpose of which is to lower the barrier to making music and to enable easier collaboration. This is done in the context of the Creativitas website, which has become a tool of the MIT 21M.080 "Introduction to Music Technology" class to learn about music technology and audio on the web, and a tool for FaMLE (the Fabulous MIT Laptop Ensemble) to use in live performances. The website allows creators to execute code within an editor code box and partake in a practice known as live coding, ultimately creating both sound and visuals. Audio is primarily created with the Tone.js interactive web audio framework, and visuals are drawn on a provided canvas using p5.js. This thesis extends the Creativitas website by providing functionality for exporting the written code as a standalone website. The exported standalone websites serve as DMIs, with standard controls such as volume, tempo, and start and stop buttons. Furthermore, we discuss and implement strategies for synchronizing timing and instrument values. This includes state-of-the-art strategies, as well as ideas for creating extendable interfaces that can include more strategies as they are developed. We end with two examples of exported DMIs, which can be effectively used in performances.

Programmable Expressiveness in Non-Social Tasks: A Mixed-Methods Study of Middle School AI Learning

Thu, 01 May 2025 00:00:00 GMT

Programmable Expressiveness in Non-Social Tasks: A Mixed-Methods Study of Middle School AI Learning Lei, Si Liang Background. Programmable expressive features—such as speech, facial expressions, and chatbot-style dialogue—are often promoted as tools to enhance engagement in educational robotics. While prior research shows benefits in socially-oriented tasks like storytelling or group collaboration, it remains unclear how student-controlled expressive blocks affect learning when the task itself is non-social. This study isolates the impact of such features in a context where expressiveness is not instructionally required. Method. We conducted a controlled, two-cohort study with 41 middle school students (ages 10–12) during a one-day AI-and-robotics workshop using the Doodlebot platform. Students in the experimental group had access to optional blocks enabling the robot to speak, emote, and use GPT-based responses. These features were hidden from the control group. All participants completed identical programming tasks (e.g., maze navigation, visual classification) that did not require social interaction. Data sources included pre/post surveys, facilitator notes, and student code. We applied the Mann–Whitney U test [1, 2] and reflexive thematic analysis [3, 4] to examine outcomes. Results. The expressive condition showed no significant gains in programming confidence or peer trust, but performed significantly worse on the post-workshop concept quiz (p = .007, r = .41). Qualitative data revealed that students in this group often used expressive blocks for entertainment rather than learning, leading to distraction, off-task behavior, and increased reliance on adult facilitation. Contributions. This study contributes (i) empirical evidence on the limitations of robot expressiveness in non-social learning contexts, (ii) a mixed-methods protocol for analyzing classroom robot deployments, and (iii) design guidance for aligning robot behavior with pedagogical intent. Implications. Expressiveness in educational robots should be contextually deployed—not assumed beneficial by default. In technical, goal-driven tasks that do not involve social reasoning, unscaffolded expressiveness may introduce cognitive overhead or divert attention. We propose a “dial-a-sociality” model, where robot behavior can be flexibly tuned to match the demands of the learning environment.

High-Speed Simulator for Millimeter-Wave Synthetic Aperture Radar

Thu, 01 May 2025 00:00:00 GMT

High-Speed Simulator for Millimeter-Wave Synthetic Aperture Radar Kuka, Adrian The past few years have witnessed growing interest in using millimeter-wave signals for non-line-of-sight (NLOS) perception tasks, with applications in robotics, augmented reality, and smart-homes. However, existing systems suffer from a lack of large mmWave datasets, resulting in limited accuracy and generalizability compared to their line-of-sight, camera-based counterparts. We present the design, implementation, and evaluation of mmSim, a new, high-speed millimeter-wave (mmWave) simulator capable of producing large synthetic datasets to help drive the field of mmWave-based NLOS perception. mmSim introduces two main contributions to improve the speed over existing mmWave simulators. First, it pre-selects areas of the object, which will produce reflections towards each simulated antenna location, allowing it to minimize future computation. Second, it introduces a coarse-to-fine approach to allow early, less critical steps to operate at lower resolutions, while maintaining the high resolution in later steps required for high-accuracy images. These techniques, combined with other performance optimizations, allow mmSim to achieve an over 24x improvement in speed over state-of-the-art mmWave simulators.

Towards AI Safety via Interpretability and Oversight

Thu, 01 May 2025 00:00:00 GMT

Towards AI Safety via Interpretability and Oversight Kantamneni, Subhash In this thesis, we advance AI safety through mechanistic interpretability and oversight methodologies across three key areas: mathematical reasoning in large language models (LLMs), the validity of sparse autoencoders, and scalable oversight. First, we reverse-engineer addition within mid-sized LLMs and discover that LLMs represent numbers as helices. We demonstrate that LLMs perform addition via the manipulation of these helices using a "Clock" algorithm, providing the first representation-level explanation of mathematical reasoning in LLMs, verified through causal interventions on model activations. Next, we rigorously evaluate sparse autoencoders (SAEs), a popular interpretability tool, by testing their effectiveness on the downstream task of probing. We test SAEs under challenging probing conditions, including data scarcity, class imbalance, label noise, and covariate shift. While SAEs occasionally outperform baseline methods, they fail to consistently enhance task performance, underscoring a potentially critical limitation of SAEs. Lastly, we introduce a quantitative framework to evaluate scalable oversight - a promising idea where weaker AI systems supervise stronger ones - as a function of model intelligence. Applying our framework to four oversight games ("Mafia," "Debate," "Backdoor Code," and "Wargames"), we identify clear scaling patterns and extend our findings through a theoretical analysis of Nested Scalable Oversight (NSO), deriving conditions for optimal oversight structures. Together, these studies advance our understanding of AI interpretability and alignment, providing insights and frameworks to progress AI safety.

Metagradient Descent: Differentiating Large-Scale Training

Thu, 01 May 2025 00:00:00 GMT

Metagradient Descent: Differentiating Large-Scale Training Chen, Benjamin A major challenge in training large-scale machine learning models is configuring the training process to maximize model performance, i.e., finding the best training setup from a vast design space. In this work, we unlock a gradient-based approach to this problem. We first introduce an algorithm for efficiently calculating metagradients -- gradients through model training -- at scale. We then introduce a "smooth model training" framework that enables effective optimization using metagradients. With metagradient descent (MGD), we greatly improve on existing dataset selection methods, outperform accuracy-degrading data poisoning attacks by an order of magnitude, and automatically find competitive learning rate schedules.

A simplified approach to calculating personalized estimates for electric vehicle charging delays

Thu, 01 May 2025 00:00:00 GMT

A simplified approach to calculating personalized estimates for electric vehicle charging delays Chen, Helen In the past decade, electric vehicles (EVs) have gained traction as a cleaner alternative to internal combustion engine vehicles, commonly referred to as gas-powered vehicles. To promote EV adoption, the government has implemented various regulations and incentives to support the transition to cleaner transportation. However, EV adoption in the United States has progressed more slowly than expected, with EVs accounting for less than 10 percent of new vehicle sales in 2023. Recent surveys indicate that a significant barrier is the perceived inconvenience and uncertainty surrounding EV charging, particularly the additional time required to charge during active use, which we call charging delay. Currently, there exist some models for estimating these charging delays, but these models require users to input a significant amount of information, such as their daily driving schedules, locations of charging stations, and exact distances of trips taken each year, which many users may not even remember. These more complex models are likely to overwhelm users, especially those who may be entirely new to EVs. To fill this gap, this thesis introduces a simplified model for estimating personalized annual EV charging delay using a set of easy-to-provide inputs, including typical driving behavior and access to home and work charging. The model logic captures delay from both routine usage, such as weekly driving patterns or typical trips, and occasional, high-energy long-distance trips, which, while not routine, are still important to account for. For weekly trips, the model considers four scenarios based on combinations of home and work charging access to determine driving and charging schedules. For long-distance travel, the model uses data from the 2022 National Household Travel Survey (NHTS) and performs multiple iterations of bootstrap resampling to create synthetic distributions of long-distance trips within a year. Data related to individual routine vehicle usage and charging delay is unavailable, so we are unable to validate the model’s performance through accuracy calculations. Instead, we performed a one-at-a-time sensitivity analysis to better understand how charging delay is affected by different factors. We found that access to private charging, such as home or work charging, improves charging delay robustness for regular weekly trips, with the exception that relying solely on work charging on workdays can cause stepwise increases in non-workday delays. Additionally, long-distance trip delays are no affected by private charging access and follow a stepwise pattern based on vehicle range. In general, the simplified approach presented in this thesis offers a more accessible way for current and prospective EV owners to clearly understand their own expected experience of EV ownership.

The Limits of Temporal Abstractions for Reinforcement Learning with Sparse Rewards

Thu, 01 May 2025 00:00:00 GMT

The Limits of Temporal Abstractions for Reinforcement Learning with Sparse Rewards Li, Zhening Skills are temporal abstractions that are intended to improve reinforcement learning (RL) performance through hierarchical RL. Despite our intuition about the properties of an environment that make skills useful, there has been little theoretical work aimed to characterize these properties precisely. This work studies the utility of skills in sparse-reward environments with a discrete state space and finite action space. We show, both theoretically and empirically, that RL performance gains from skills are worse in environments where successful trajectories are less compressible. In environments with a highly incompressible distribution of successful trajectories, using unexpressive skills such as macroactions will provably worsen RL performance. We hope our findings can guide research on automatic skill discovery and help RL practitioners better decide when and how to use skills.

Self-Supervised ECG Learning for Multimodal Clinical Tasks

Thu, 01 May 2025 00:00:00 GMT

Self-Supervised ECG Learning for Multimodal Clinical Tasks Chen, Peilin We present a multimodal clinical AI framework that integrates time series, images, and text to support robust diagnostic reasoning across diverse input combinations. We first introduce ECG-JEPA, a self-supervised encoder pretrained on multiple ECG datasets to learn generalizable time series representations. This unimodal pretraining improves ECG classification, achieving a 23-point AUC gain on the underrepresented Ga dataset. We then align and fuse these ECG embeddings with chest X-rays and EHR text using a vision–language model backbone, enabling end-to-end multimodal inference. Our results show that incorporating ECG signals meaningfully improves diagnostic performance, highlighting the value of multitask time series pretraining and modular fusion for clinical AI.

A Hierarchical Approach to Quantitative Portfolio Optimization for Technology Development Project Portfolios (OPTIM-H)

Thu, 01 May 2025 00:00:00 GMT

A Hierarchical Approach to Quantitative Portfolio Optimization for Technology Development Project Portfolios (OPTIM-H) Huang, Roderick W. The use of Mean-Variance Portfolio Optimization (MVO) in Modern Portfolio Theory (MPT) has been a long-standing method to guide investment decisions for market-traded assets like stocks and bonds. Recent research shows that portfolio optimization developed using MPT could prove useful in investment decisions for technology projects. Traditionally, empirical data from past projects and statistically driven technology trends are used to predict the risk-return model necessary for MPT. This thesis introduces a new methodology, Optimizing Portfolios in Technologies Investments Methodology with Hierarchy (OPTIM-H), which extends MPT to make investment decisions within a hierarchical organizational structure of technology projects. An integrated dataset was developed to demonstrate this methodology, combining 19,000 data records from Techport and Small Business Innovation Research (SBIR) datasets. The dataset captures investment trends and maturity pathways across 17 taxonomy areas, revealing that most projects begin at Technology Readiness Levels (TRLs) 2–4, with average funding amounts near \$300,000. OPTIM-H effectively distinguishes between broader technology groups and their subcategories, showing the impact of community interest on investment decisions. Furthermore, this work investigates k-means clustering as a tool for classifying technology projects for targeted investment, with the analysis identifying seven clusters and achieving a mean utility score of 0.595 with a standard deviation of 0.651.

Integrating Canvas with a Large-Scale Social Annotation Platform

Thu, 01 May 2025 00:00:00 GMT

Integrating Canvas with a Large-Scale Social Annotation Platform Heiberger, Henry R. The last decade has seen a growing interest in the use of collaborative annotation systems, educational tools that allow multiple users to asynchronously comment, highlight, and discuss digital content directly on the source material, transforming traditional classroom readings into a more engaging group activity. Originally developed by MIT CSAIL’s Haystack Group in 2012 under the direction of Professor David Karger, Nota Bene (NB) is a particular collaborative annotation tool that allows students to have annotated online discussions in the margins of textbooks, papers, and even webpages [1]. Though various studies have already proven its ability to succeed in a classroom setting, conversations with key stakeholders have revealed that the tool is missing a key feature found in many other popular collaborative annotation solutions: integration with the Canvas learning management system (LMS) [1–3]. Thus, this work sought to integrate the classroom management features that Canvas provides into the NB platform by supporting Canvas account linking, class importation and roster synchronization, and automatic grade uploading. By doing this, we hoped to improve NB’s quality as a classroom tool, enhancing its value to institutions, encouraging its wider adoption across the academic landscape, and aligning with a much broader trend of creating more integrated, efficient, and user-friendly educational technology solutions.

On Passive-Scoping as a method for Large Language Model Robustness to Jailbreaks and Adversarial Examples

Thu, 01 May 2025 00:00:00 GMT

On Passive-Scoping as a method for Large Language Model Robustness to Jailbreaks and Adversarial Examples Hernandez, Adriano Artificial Intelligence (AI) and large language models (LLMs) not only present a challenge for adversarial robustness, but also the natural emergence of unwanted capabilities. Current approaches to safeguarding AI and LLMs predominantly rely on explicitly restricting known instances of these. However, this places a burden on model developers, because they cannot anticipate all the potential attacks and undesirable capabilities. To solve this problem, we leverage interdisciplinary knowledge. In the field of information security, the principle of least privilege provides guidance on how to defend from unknown threats. In AI, the principle could be implemented by ensuring that developers specify the knowledge and capabilities an AI system should retain, restricting all others by default. We call this application of the principle of least privilege, passive scoping. Our thesis makes two claims: 1. We argue that (a) passive scoping mitigates concerns about adversarial robustness and loss of control of AI systems and (b) passive scoping to edit the weights and activations at post-training time is underexplored by the literature. 2. Of possible approaches, our sparse autoencoder (SAE) filters can implement this underexplored type of passive scoping. They increase safety relative to LoRA finetuning and prompt engineering, but leave room for improvements. The thesis is structured as follows: 1. Chapter 2 elucidates the challenges with adversarial robustness and loss of control risk. Chapter 3 puts forward a conceptual argument for the benefits of passive scoping. Later, it analyzes the extent to which passive scoping has been attempted. These two chapters work together to defend claims 1a and 1b. 2. Chapter 4 defines our optimization problem. Chapter 5 defines our experimental methodology and metrics. These two define our success criteria for claim 2. Chapter 6 finalizes our defense of claim 2 based on our results. 3. Chapter 7 explores related work, Chapter 8 engages in a broader discussion, and chapter 9 summarizes the contributions of this thesis.

Explorations in AI and Creative Learning New Tools to Expand How Young People Imagine, Create, and Tinker with Scratch

Thu, 01 May 2025 00:00:00 GMT

Explorations in AI and Creative Learning New Tools to Expand How Young People Imagine, Create, and Tinker with Scratch Huang, Alexis As generative AI tools become increasingly prevalent in young people’s lives, these technologies have a growing influence over the way that children learn. While much of the early work at the intersection of AI and education has focused on the development of intelligent tutoring systems designed to deliver content more efficiently, this thesis explores how generative AI might be used to support the creative learning process by sparking curiosity, encouraging exploration, and helping young people express themselves creatively. In this thesis, I explore ways of integrating generative AI with Scratch, the world's largest programming community for children, while remaining aligned with the core values of Scratch: creativity, playfulness, and self-expression. I designed three tools that extend the Scratch ecosystem: Scratch Connect, which explores using generative AI to help Scratchers discover projects that inspire them to create while opening the black box of recommendation systems; scrAItch, which investigates how people can iterate with generative AI by using text-based inputs to create and tinker with Scratch projects; and Scratch Spark, which reimagines the new learner experience by using generative AI to help users create personally meaningful “spark projects.” This thesis describes the process of imagining, creating, and reflecting on these tools, including many of the challenges and tensions that we encountered along the way. I discuss observations and feedback from creative workshops with young people, and conclude by reflecting on open questions and opportunities for future work in designing generative AI tools that support creative learning.

Effects of Hardware Design Choices on Neural Network Accuracy in Analog Inference Accelerators

Thu, 01 May 2025 00:00:00 GMT

Effects of Hardware Design Choices on Neural Network Accuracy in Analog Inference Accelerators Forsythe, Eyan Analog accelerators can enable energy-efficient and high-throughput deep neural network (DNN) computations by computing in memory. Unfortunately, device and circuit nonidealities in these accelerators, such as noise and quantization, can also lead to low DNN inference accuracy due to computation errors arising from these non-idealities. These errors are largely a function of both the choice of DNN workload and different hardware design choices, such as circuit topology and DNN operand encoding. Different hardware design choices can affect the energy, throughput, and area of the system, so it is important to understand how these design choices interact with DNN inference accuracy. However, there is a lack of a systemic understanding of how each of these hardware design decisions affects accuracy and how they interact with other design decisions. To address these issues, we model how hardware design choices can lead to analog errors such as noise and quantization. Then, we explore these errors affect inference accuracy in analog accelerators and how tradeoffs can be made between inference accuracy, energy efficiency, area, and throughput. We find that analog errors generated from hardware design decisions can generate different amounts of accuracy loss depending on which layer in a DNN is subject to these analog errors. This leads to the structure of the DNN having a significant impact in how hardware design choices affect DNN inference accuracy, especially with respect to the individual layers of a DNN. We use knowledge of the relationships between device and circuit non-idealities to improve the accuracy of published analog accelerators and analyze the energy and area costs of the increased accuracy.

Design of High-Resolution SAR ADC for Detection of Sub-Cortical Neuron Action Potentials for BMI Applications

Thu, 01 May 2025 00:00:00 GMT

Design of High-Resolution SAR ADC for Detection of Sub-Cortical Neuron Action Potentials for BMI Applications Guobadia, Omozusi E. The advancement of brain-machine interfaces (BMIs) requires neural signal acquisition systems that are capable of resolving both fast, low-amplitude action potentials (APs) and slow, higher-amplitude local field potentials (LFPs) under stringent power and area constraints. This thesis presents the design and simulation of a high-resolution, low-power successive approximation register (SAR) analog-to-digital converter (ADC) tailored for sub-cortical neural signal detection. To optimize dynamic range and reduce power consumption, a novel adaptive zoom-and-tracking architecture is introduced, enabling the ADC to dynamically adjust its reference window based on LFP trends while maintaining high-resolution capture of APs. The proposed system integrates a bootstrapped track-and-hold circuit, a differential capacitive DAC, and a strong-arm comparator in the analog front-end, alongside a digital FIR filter and SAR logic with zoom-range control in the digital domain. Simulations validate the functionality of each subsystem independently and in concert, demonstrating the system’s ability to dynamically isolate APs from LFP-dominated baselines while reducing analog power draw by over 60% compared to fixed-range ADCs. This work offers a promising approach for scalable, energy-efficient neural recording architectures suited to future BMI applications.

Transformer-Based Prediction of Coronary Artery Lumen Expansion Post Angioplasty Using Optical Coherence Tomography

Thu, 01 May 2025 00:00:00 GMT

Transformer-Based Prediction of Coronary Artery Lumen Expansion Post Angioplasty Using Optical Coherence Tomography Gupta, Shreya Coronary artery disease is the leading cause of mortality globally, resulting in an urgent and critical need to better understand both vessel morphology and the processes of intervention. Angioplasty is an intervention which causes a previously constricted vessel to expand via placement of a stent, and is affected by numerous characteristics of the vessel such as calcium eccentricity and size, wall thickness, and prior lumen size. Being able to accurately assess whether a stent will properly expand allows cardiologists to pursue pre-stenting calcium lesion modification strategies that help avoid dangerous complications of improper stenting. This work introduces a pipeline for post-stenting lumen area prediction from pre-stenting optical coherence tomography (OCT) images. This pipeline includes morphological correction of OCT image segmentations, explainable feature extraction from OCT segmentations, and a predictive transformer network that combines morphological features with injected stent information. The aim is for such a pipeline to be used to support clinical decision making.

Complete Visual and Geometric Object Reconstruction via Autonomous Robotic Manipulation

Thu, 01 May 2025 00:00:00 GMT

Complete Visual and Geometric Object Reconstruction via Autonomous Robotic Manipulation Fu, Evelyn Accurately simulating object dynamics based on real-world perception inputs has wide applications in digital twins and robotic manipulation. Yet, doing so requires practitioners to carefully measure and reconstruct the dynamic and geometric properties of the objects, which is time-consuming and requires domain expertise. This project proposes an automatic pipeline to construct 3D representations from a collection of real objects, which can further be used to generate assets with accurate visual texture and collision geometry to be used in simulation. This pipeline will be designed to have minimal hardware requirements and aim to be efficient in time for physical actuation to maximize data collection on minimal hardware.

Model-based Planning for Efficient Task Execution

Thu, 01 May 2025 00:00:00 GMT

Model-based Planning for Efficient Task Execution Ding, Wenqi Robotic agents navigating 3D environments must continuously decide their next moves by reasoning about both visual observations and high-level language instructions. However, they plan in a high-dimensional latent space, opaque to human collaborators. Hence, it is difficult for humans to understand the agent’s decision-making process. This lack of interpretability hinders effective collaboration between humans and robots. The key question we are trying to answer in this thesis is: Can we build a unified planning framework that fuses visual and language into a single, interpretable representation, so that humans can interpret robots’ decisions? We propose a model-based planning framework built around pretrained vision-language models (VLMs). We show that VLMs can be used to plan in a unified embedding space, where visual and language representations can be decoded back to human-interpretable forms. Empirical evaluation on vision-language navigation benchmarks demonstrates both improved sample efficiency and transparent decision making, enabling human-in-the-loop planning and more effective human-robot collaboration.

Global Non-Convex Optimization with Integer Variables

Thu, 01 May 2025 00:00:00 GMT

Global Non-Convex Optimization with Integer Variables Kriezis, Demetrios C. Non-convex optimization refers to the process of solving problems whose objective or constraints are non-convex. Historically, this type of problems have been very difficult to solve to global optimality, with traditional solvers often relying on approximate solutions. Bertsimas et al. [1] introduce a novel approach for solving continuous non-convex optimization problems to provable optimality, called the Relaxation Perspectification Technique - Branch and Bound (RPT-BB). In this thesis, we extend the RPT-BB approach to the binary, mixed-binary, integer, and mixed-integer variable domains. We outline a novel branch-and-bound algorithm that makes use of the Relaxation Perspectification Technique (RPT), as well as binary, integer, and eigenvector cuts. We demonstrate the performance of this approach on two representative non-convex problems, as well as two real-world non-convex optimization problems, and we benchmark its performance on BARON and SCIP, two state-of-the-art optimization solvers for non-convex mixed-integer problems. We observe that our algorithm, despite being more general, is able to outperform the state-of-the-art solvers on many problem instances.

Optimizing AI Agents for Automated Software Engineering with Palimpzest

Thu, 01 May 2025 00:00:00 GMT

Optimizing AI Agents for Automated Software Engineering with Palimpzest Li, Jason The deployment of large language models (LLMs) as autonomous agents is transforming the software development landscape. Increasingly more engineers are using natural language agents to expedite and guide development workflows, while large organizations are investing heavily on building agentic systems for tasks such as code generation and code repair. A key challenge in developing such systems is tuning agent hyperparameters— settings that affect performance such as choice of model, temperature settings, and context window sizes. As system complexity grows, the hyperparameter space expands, complicating optimization under real-world compute and time constraints. In this work, we present Palimpzest[1] as an agentic optimizer able to balance cost and performance objectives by tuning agentic hyperparameters. We demonstrate that Palimpzest can tune our agent hyperparameters at 8.5 times lower cost and with 24 times greater time efficiency compared to the conventional grid search. By integrating our custom-built Debugger and Code Editor Agents as new operators within Palimpzest, we enhance the system’s ability to resolve real-world GitHub issues. And to facilitate hyperparameter selection, we also introduce File Coverage, Report Accuracy, and Patch Similarity along with the traditional SWE-Bench Score as quality evaluation methods used by Palimpzest’s optimization loop. When evaluated on the SWE-Bench Lite[2] benchmark, our optimized system achieves a 15% score at a significantly lower cost compared to previous approaches.

Integrating Gradient Boosting and Generative Models: Hybrid Approach to Address Class Imbalance and Evaluation Gaps in Real-World Systems

Thu, 01 May 2025 00:00:00 GMT

Integrating Gradient Boosting and Generative Models: Hybrid Approach to Address Class Imbalance and Evaluation Gaps in Real-World Systems Lau, Mary Anomaly detection remains a persistent challenge in machine learning due to the extreme class imbalance, high cost of false negatives, and the need to regulate false positives in realworld settings at scale. This thesis introduces Tail-end FPR Max Recall, a business-aware evaluation framework designed for such constrained environments. Using this framework, we benchmark LightGBM—a gradient boosting method known for its computational efficiency and predictive accuracy—on an imbalanced dataset, comparing its performance against standard academic evaluation criteria. Our results demonstrate that Tail-end FPR Max Recall fills critical gaps left by standard academic criteria, providing a more realistic assessment of model performance that aims to maximize recall while enforcing a false positive rate budget. Beyond benchmarking, we propose two strategies that incorporate deep learning methods to augment the already strong performance of gradient boosting: (1) using generative models to produce synthetic minority-class samples that outperform traditional oversampling techniques, and (2) using neural embeddings to improve feature representation for anomaly detection. Together, these contributions offer a methodology for evaluating and improving anomaly detection pipelines in domains where rare, high-impact events must be detected while meeting strict operational demands.

FPGA Based Data Acquisition System for Cryogenic Device Verification

Thu, 01 May 2025 00:00:00 GMT

FPGA Based Data Acquisition System for Cryogenic Device Verification Kandeh, Stephen In this work, a system of processors connected to an FPGA is interfaced with a custom analog frontend and used to create a verification environment for cryogenic devices. In particular, this thesis focuses on the technical structure of that system. Current validation efforts often rely on commercially available arbitrary waveform generators (AWGs) and oscilloscopes, which, while highly capable, are often prohibitively expensive and poorly suited for large-scale or parallelized testing environments. As noted in industry reports, scaling such instrumentation introduces significant challenges in cost, calibration, and signal synchronization, making them inefficient for high-resolution or high-speed analyses in multi-channel systems [1]. On the other hand, an FPGA provides the necessary performance to increase parallelism without a proportional increase in cost, greatly improving testing resolution and speed. When augmented with a set of processors, we introduce a level of accessibility and automatability not currently present in commercial products. To be clear, while the board was designed with the testing of nanowires in mind (and is not capable of measuring DC voltages), it can still be combined with separate lab equipment to interact with Josephson Junction based devices. That said, the flexibility of this system allows for a generalized application to any electronic that demands a specialized testing procedure involving arbitrary signal processing and generation. The money, time, and energy that this innovation will save on cryogenic electronic validation will significantly improve our progress in developing these technologies.

Energy Efficient Real-time Operating Systems on Chip

Thu, 01 May 2025 00:00:00 GMT

Energy Efficient Real-time Operating Systems on Chip Kang, Ezra H. Autonomous micro-robots are crucial for several tasks, such as search and rescue, noknowledge mapping, and navigation. Without an external power connection, these robots are constrained by their on-platform energy capacity. The power consumption of actuation systems used in micro-robots is within the same magnitude of the power consumption of the compute system. Thus, the remaining factor for enabling these micro-robots is associated with the design of energy-efficient compute systems. Energy usage of compute systems is typically dominated by memory operations, which previous efforts have attempted to mitigate with memory efficient software and hardware. These efforts are enabled with the software/hardware interface, which is implemented as an Operating System (OS). However, Operating Systems for energy-efficient platforms have not been fully explored. Current approaches utilize full general-purpose Operating Systems such as Linux, which can incur large memory and compute overhead penalties. These overheads not only consume the typically limited memory resources of energy-efficient systems, but also increase the number of memory accesses and CPU cycles, both of which are significant contributors to energy consumption. To address these concerns, we propose the design of a computational and memory efficient Real-time Operating System (RTOS). Our RTOS is designed to minimize both memory footprint and compute cycle overhead. It achieves this primarily through direct physical memory access, cycle-efficient task scheduling, and minimal runtime services to avoid unnecessary processing. Additionally, the modular RTOS kernel includes only the components required by an application in the final binary, reducing code size and memory usage without compromising functionality. The design enables the utilization of energy-efficient hardware accelerators and software, allowing for execution of robotics workloads with minimal memory and cycle overhead. When comparing robotics algorithms implemented on our proposed RTOS and baseline OSes, our design was able to achieve a 99% reduction in memory footprint. Additionally, it achieved up to a 47% increase in throughput. Thus, our design demonstrates a direct reduction in memory and CPU cycle overhead, which in turn lowers total system memory and energy consumption. The proposed design was demonstrated and verified on a resource constrained system-on-chip on the AMD Virtex Ultrascale+ VCU118 FPGA.

Uncertainty and Generality of Transfer Learning Models in Predicting Signaling History

Thu, 01 May 2025 00:00:00 GMT

Uncertainty and Generality of Transfer Learning Models in Predicting Signaling History Lu, Claire Proper cell-cell communication is essential for multicellular development, from embryogenesis to stem cell differentiation. To map these networks, we developed IRIS (Intracellular Response to Infer Signaling state), a semi-supervised deep learning method that fits conditional variational autoencoders (CVAE) to single-cell RNA sequencing (scRNA-seq) data. IRIS is able to annotate cellular signaling states of individual cells using only their gene expression. Currently, IRIS has been validated in developmental contexts, including gastrulation, early endoderm organogenesis, and mesoderm lineages in mouse embryos. However, its predictions often show extremely high or extremely low confidence, suggesting a need for methods to prevent overconfidence and better account for uncertainty. To generalize IRIS to broader cell-cell communication problems, we combined engineering and experimental approaches, integrating uncertainty quantification techniques with new biological datasets. We implemented three approaches for estimating uncertainty in IRIS predictions: stochastic sampling, Monte Carlo dropout, and ensemble prediction. These approaches were evaluated on two new endoderm and mesenchyme combinatorial perturbation screens. Across all methods, uncertainty values reliably reflected the varying difficulty of predicting different signaling pathways, driven by both biological complexity and dataset representation. Moreover, higher uncertainty was consistently associated with lower prediction accuracy, confirming uncertainty as a useful proxy for model confidence. All three methods identified similar high-uncertainty cell populations, supporting their consistency and validity. By incorporating uncertainty quantification into IRIS, we provide more robust and interpretable predictions that can guide future experiments and enhance the model’s applicability across diverse biological contexts.

Core Material Evaluation for Magnetic Energy Harvester Applications

Thu, 01 May 2025 00:00:00 GMT

Core Material Evaluation for Magnetic Energy Harvester Applications Le, Khang D. Current transformer magnetic energy harvesters (CTMEHs) harvest magnetic energy from an AC current-carrying conductor and convert this energy into usable electrical energy for use by various low-power devices, such as sensors and microcontrollers. The amount of power harvested by CTMEHs is determined by the primary current passing through the conductor; however, variables such as the magnetic core’s dimensions, magnetic properties, and turn count also influence performance. Previous works have focused mainly on analytical or numerical modeling of CTMEH behavior or improving power harvest performance given a specific magnetic core material. Some existing research has compared the effects of different core materials on CTMEH power harvest in limited fashion; but a comprehensive, comparative study of high permeability, high saturation flux density CTMEHs had yet to be explored. This thesis establishes core material as the primary independent variable along with primary current and frequency during testing to isolate the effects of magnetic properties on determining the amount of power a magnetic core can harvest under different current conditions. The thesis concludes that nanocrystalline material excels at lower-current applications, while silicon steel material offers better performance at higher-current applications across all frequencies when used as CTMEHs, offering system designers enticing material choices depending on the nature of the application.

Eliciting Visualization Attitudes with Repertory Grids

Thu, 01 May 2025 00:00:00 GMT

Eliciting Visualization Attitudes with Repertory Grids Hua, Dana Research in public data communication typically focuses on improving the processes of encoding and decoding, answering the question of how to design a visualization to best communicate information to an audience. However, by treating visual communications as simply conduits for information, we ignore an important aspect of how people interact with communications. We ignore the attitudes – the thoughts, feelings, and intentions toward action – a person may form from communicative artifacts based on their personal values and experiences. Recent research has demonstrated that—much like natural language—readers of visualizations make social attributions: inferences about the identities and characteristics of an artifact’s makers, modes of distribution, and tools of production. In this thesis, I contribute a method to systematically map the visualization attitudes of an individual and the associated ideologies of their sociocultural group, by adapting the repertory grid technique from clinical psychology, to the context of data visualization. I demonstrate the effectiveness of this mixed methods approach by eliciting both the attitudes towards a visualization most salient to an individual, and the design features of the visualization that inform each attitude. This method offers a new way of exploring the content and latent structure of visualization attitudes, which opens new avenues for socioculturally-informed and intervention-driven research in data visualization.

Optimizing scheduling for stream structured programming for StreamIt

Thu, 01 May 2025 00:00:00 GMT

Optimizing scheduling for stream structured programming for StreamIt Dow, Nicholas Lee As straightforward increases in performance on general purpose CPUs slow down, the shift to application specific implementations and hardware has accelerated. This shift to towards specialization improves performance but often at the cost of developer productivity in learning these new tools. StreamIt is a Domain Specific Language developed to increase performance of streaming applications while being relatively user-friendly. While designed to be parallelized easily, the scheduling backend of the StreamIt compiler is not adapted to the heterogeneous and distributed nature of new accelerator hardware. This thesis details the design and development of a scheduler interface that enables hardware customized schedulers to be developed quickly. The scheduler interface allows for schedulers to take advantage of the unique compiler optimizations enabled by StreamIt’s structure. Two schedulers, one search based and another heuristic based, are built using this interface to schedule StreamIt workloads to optimize differing metrics such as throughput and latency. Our experiments evaluate the performance of these workloads, and details future direction for expanding the interface and scheduler designs that could take advantage of it.

Mitigating Electromagnetic Interference in Unshielded MRI: Implementation, Experimentation, and Future Directions

Thu, 01 May 2025 00:00:00 GMT

Mitigating Electromagnetic Interference in Unshielded MRI: Implementation, Experimentation, and Future Directions Flynn, John M. Portable, Low-Field MRI broadens access and enables numerous new applications such as point-of-care. Operating outside an RF-shielded room introduces electromagnetic interference (EMI), degrading further the signal-to-noise ratio (SNR) which is already diminished due to the lower magnetic fields used in portable imaging. Existing methods to reduce EMI perform well in simple noise environments, but can struggle with more complex profiles. Relaxing the linear assumptions is hypothesized to bring more robust mitigation algorithms. A system-wide characterization of SNR challenges was carried out on a rebuilt 800G scanner, existing techniques were validated, and new signal processing approaches were explored to drive image quality upwards. Various analytical approaches showed promise, such as dynamic coils/preamps, averaging methods, calibration, and smoothing methods. Groundwork was laid for learning-based methods throughout the pipeline. This work serves as an important baseline for the numerous experiments necessary for the full-system optimization.

Exploring Atom-Light Scattering in the Quantum Regime

Thu, 01 May 2025 00:00:00 GMT

Exploring Atom-Light Scattering in the Quantum Regime Lu, Yu-Kun Ultracold atoms and molecules are promising platforms for exploring modern quantum science and technologies, such as quantum simulation and quantum computation. Here, light is the essential tool to manipulate and probe these systems. However, unlike in condensed matter systems where scattering experiments are routinely employed to characterize materials, ultracold atom and molecule systems are usually probed by imaging and not by light scattering. In this thesis, I present a systematic investigation of atom-light scattering under various scenarios. When atoms are confined in optical lattices, light scattering can be used to explore single-body, two-body, and many-body physics. Focusing on single-atom physics, I study coherent and incoherent light scattering of single-atom wavepackets and the relation to which-way information. For two atoms tightly localized to a 20nm size on the same lattice site, I demonstrate the strong electric dipolar interactions between them, which result in large momentum transfers and spectroscopic shift of the resonance. On the many-body side, I show how light scattering can reveal distinct quantum phases at thermal equilibrium or defect generation in dynamical ramps. For atoms released from the optical lattice, I demonstrate that light scattering can read out the quantum statistical information and initial density correlations hidden in the interference of atomic wavepackets. When atoms move freely in the form of degenerate quantum gases, I investigate how quantum statistics, phase transition, and interactions modify the atomic pair correlation and consequently the light scattering. For thermal gas at high density, I demonstrate nonlinear optical effects from high optical density and high scattering rate. Finally, I describe our recent efforts on manipulating atoms at subwavelength length scales. I discuss our attempts in optical tweezers and in optical lattices, and the prospect of observing magnetic pairing between two distant layers under attractive dipolar interaction. The techniques presented in this thesis should be of general use for pursuing quantum science and technology with ultracold atoms and molecules.

Single-Model Any-Subgroup Equivariance via Symmetric Positional Encodings

Thu, 01 May 2025 00:00:00 GMT

Single-Model Any-Subgroup Equivariance via Symmetric Positional Encodings Goel, Abhinav The inclusion of symmetries as an inductive bias, known as “equivariance”, often improves generalization on geometric data (e.g. grids, sets, and graphs). However, equivariant architectures are usually highly constrained, designed for pre-chosen symmetries, and cannot be applied to datasets with different symmetries. This work constructs a single model that is simultaneously equivariant to several groups, by simply regulating a certain input feature. Starting with a permutation-equivariant base model respecting the full Sₙ symmetry group, we can obtain subgroup G ⊆ Sₙ equivariance by using a symmetry-breaking input that is G-symmetric. Under mild conditions, the resultant network is only G-equivariant. But finding an input with automorphism group exactly G is computationally hard, which can be overcome by relaxing exact symmetry breaking to approximate symmetry breaking. This is done by leveraging the notion of 2-closure to derive fast algorithms. This method is validated on symmetry selection, multitask, and transfer learning settings, demonstrating that a single network equivariant to multiple permutation subgroups outperforms both separate equivariant models or a single non-equivariant model.

Application of Precision Successive-approximation-register Analog-to-digital Converters for Digital Root-mean-square Calculation

Thu, 01 May 2025 00:00:00 GMT

Application of Precision Successive-approximation-register Analog-to-digital Converters for Digital Root-mean-square Calculation Choi, Sun Mee The advancement of semiconductor manufacturing processes has allowed for the availability of powerful microcontrollers at lower costs, granting system designers the flexibility to select between analog and digital signal processing techniques. Enabled by recent developments in low-power successive approximation register (SAR) analog-to-digital converter (ADC) technology, a digital approach to root-mean-square (RMS) measurement is proposed. The work begins with an explicit accumulation and averaging approach, and a set of improvements were designed to increase measurement accuracy and reliability. Algorithms are compared using the metrics of error, power efficiency, latency, and digital overhead. High-performing and power-efficient digital RMS measurement methods could be valuable for decentralized instrumentation systems such as smart grids and factory automation where long-lasting handheld and portable solutions are becoming critical.

Hosting LLMs on Shared GPUs

Thu, 01 May 2025 00:00:00 GMT

Hosting LLMs on Shared GPUs Choi, Kenneth K. Large language models (LLMs) have emerged as powerful tools for a wide array of applications. Serving multiple LLMs on shared GPUs has increasingly gained attention as single providers need to support multiple applications (summarization, chat, code generation), different model versions (A/B testing), and various types of customers. However, multi-model serving is particularly challenging, as static memory partitioning can lead to severe under-utilization, fragmentation, and latency spikes, while dynamic loading of model weights can cause unacceptable downtime due to high model loading overheads. To address these issues, we introduce hierarchical paging, a novel key-value (KV) cache management strategy, and we implement it within the vLLM serving engine. Hierarchical paging organizes GPU memory into a two-level hierarchy: large contiguous memory blocks allocated to individual models, which are then subdivided into smaller blocks that are allocated to different requests issued to that model. Our design enables dynamic memory sharing across models, improving model throughput and overcoming key problems of existing approaches. We detail our implementation and present end-to-end experiments that showcase these throughput improvements under different workloads. We include further evaluations on the runtime overheads of our hierarchical paging implementation, which show that the overheads are insignificant. Most importantly, we demonstrate that hierarchical paging is easy to implement, optimizing for implementation effort and maintainability.

A Topology-Guided Diffusion Process for Synthetic Tabular Data Generation

Thu, 01 May 2025 00:00:00 GMT

A Topology-Guided Diffusion Process for Synthetic Tabular Data Generation Cheng, Emily Synthesizing realistic tabular data is crucial for any analytical application, including policy evaluation related to household energy use. However, detailed household-level consumption data, necessary for such evaluation, are scare at fine geographic scales, as public surveys like the U.S. Residential Energy Consumption Survey (RECS) provide too few observations. We address this gap by developing a topology-guided diffusion-based generative model that produces realistic synthetic household data, and our approach handles two key challenges in this setting: (1) mixed continuous and discrete features and (2) strong hierarchical dependencies among variables. To handle categorical features, we build upon recent advancements in discrete diffusion, particularly TabDDPM [1] and TabDiff [2], which discretize the diffusion process through noise transition matrices, effectively extending diffusion methods to discrete tabular domains. To address hierarchical dependence, we include (1) a structure-aware noise schedule that injects noise from the leaves to the root along an approximate Chow–Liu tree constructed from the variables and (ii) a masked self-attention denoiser that aligns with the same graphical structure. Extensive experiments show that our structured diffusion model outperforms the baseline TabDiff on data with tree-like dependencies, due to the inductive bias from our structure-aware noise schedule. On data that only approximately follows a tree, such as the RECS dataset, our model maintains competitive performance, only slightly outperforming standard diffusion methods. These results highlight the potential for future work to further optimize the tradeoff between structural approximation and estimation accuracy and for future work beyond the energy domain.

On Dynamic Treatment Regimes: Collaborative Search and LLM-Driven Decision Trees

Thu, 01 May 2025 00:00:00 GMT

On Dynamic Treatment Regimes: Collaborative Search and LLM-Driven Decision Trees Gregory, Cale This thesis evaluates the validity of current dynamic treatment regime algorithms and presents a novel data structure for extracting treatment decisions from unstructured clinical notes. The main contribution is the Clinical Decision Tree (CDT) which uses large language models (LLMs) to extract key decisions in chronic disease treatment. This addresses the main pain points in dynamic treatment regimes of low interpretability and reliance on poorly collected data for traditional machine learning methods. This work contains extensive experiments on mortality prediction, time series forecasting, and synthetic patient modeling. Experiments show that vital-based representations do not capture enough meaningful data about a patient to accurately predict and evaluate new treatment methods. By utilizing latent embeddings and vector search, experiments show that the collected vitals of patients fail to differentiate the outcomes of the related patients. Conversely, the clinical notes contain complex and substantial information about clinical decision making. LLMs enable the valuable knowledge extraction from unstructured data. Utilizing LLMs, experimental results and expert evaluation indicates that CDTs can extract and distill interpretable treatment decisions. Thus, CDTs are a valuable tool that can be refined to increase confidence in treatment decisions and identifying rare and uncommon medical practices.

"Eliminating the Friction": An AI-powered Assistant for StarLogo Nova

Thu, 01 May 2025 00:00:00 GMT

"Eliminating the Friction": An AI-powered Assistant for StarLogo Nova Han, Aileen Agent-based modeling is a technique that allows students to reason about and create models of real-life phenomena. However, the programmatic implementations of this technique, such as StarLogo Nova, often introduce “friction”; students may get stuck on the syntactical details of the implementation before being able to engage in the mechanistic thinking behind their models. In order to shift students’ focus towards the goal of understanding the systems they are building, we set out to create an AI-powered assistant for StarLogo Nova that can explain and debug students’ code. After identifying and experimenting with various parameters of AI models in an attempt to improve their performance, we were able to build the StarLogo Turtle Helper, an easily accessible assistant integrated into the platform that can produce accurate responses to StarLogo-related questions. Through this process, we discovered two key properties of these models: first, the method through which these models use provided documentation (called retrieval-augmented generation, or RAG) is quite rudimentary, so any background knowledge should be included in the prompt or the model’s system instructions instead. Second, these models perform best if they are designed to only serve one purpose, so creating multiple models and chaining them together may be the best way to achieve more complex functionality.

Predicting Progression of Metabolic Dysfunction-associated Steatotic Liver Disease

Thu, 01 May 2025 00:00:00 GMT

Predicting Progression of Metabolic Dysfunction-associated Steatotic Liver Disease Li, Jonathan This work focuses on the progression from metabolic dysfunction-associated fatty liver to metabolic dysfunction-associated steatohepatitis, a more serious prognosis that can lead to liver failure and death. Additional adverse progressed outcomes include hepatic failure, fibrosis, cirrhosis, and malignant neoplasm of liver and intrahepatic bile ducts. We explore the possibility of using different machine learning techniques, including logistic regression, XGBoost, random forest, and decision trees to predict the likelihood of progression. We use data from Massachusetts General Brigham to train our models, incorporating demographics, physical measurements, lab results, and doctor notes. As a result of this project, we our best model was an XGBoost classifier with an AUROC of 0.800 with random forest at a similar performance of 0.786. However, all of our models had low AUPRC and sensitivity, indicating both overfitting and an imbalanced dataset.

Data Traceability via OTrace Concepts and Implementation

Thu, 01 May 2025 00:00:00 GMT

Data Traceability via OTrace Concepts and Implementation Farooq, Ashar Financial transactions are commonplace in the modern world. Everyday consumers make purchases on many e-commerce sites and often use many third-party financial services, such as to predict your credit score, to obtain customized budget recommendations, and to find out which specific loan is the best for them. These financial services often need financial information from the consumer, which is not always clear to the consumer. In other words, consumer data are being used without their knowledge and consent. The proposed solution of using a traceability protocol called OTrace aims to mitigate this issue of not knowing where a consumer’s data is along with what is being done with it. This paper will aim to bolster OTrace to be more representative of a protocol that consumers can actually use as a service, and financial institutions can have trust that this will solve the problem of consumers not knowing which third-party financial services have their data. In other words, this work will create a more general traceable and accountable data sharing system specification that includes the OTrace layer on top of an OAuth layer that will be complemented with a model deployment example. The addition of more relevant OTrace API endpoints corresponding to a new specification along with an entire new OTrace Web implementation along with analysis will guide the data traceability world, data privacy world, open banking world, financial world and ultimately the global world forward. There will be a model deployment of an OTrace service on top of an OAuth protocol that can allow everyone to see it being used by various parties that can ultimately scale up to fix the problem of unintended data usage and lack of transparency of location of data.

Equivariant Autoregressive Models for Molecular Generation

Thu, 01 May 2025 00:00:00 GMT

Equivariant Autoregressive Models for Molecular Generation Kim, Song Eun In-silico generation of diverse molecular structures has emerged as a promising method to navigate the complex chemical landscape, with direct applications to inverse material design and drug discovery. However, 3D molecular structure generation comes with several unique challenges; generated structures must be invariant under rotations and translations in 3D space, and must satisfy basic chemical bonding rules. Recently, E(3)-equivariant neural networks that utilize higher-order rotationally-equivariant features have shown improved performance on a wide range of atomistic tasks, including structure generation. Previously, we have developed Symphony, an E(3)-equivariant autoregressive generative model for 3D structures of small molecules. At each sampling iteration, a single focus atom is selected, which is then used to decide on the next atom’s position within its neighborhood. Symphony built on previous autoregressive models by using message-passing with higher-order equivariant features, allowing a novel representation of probability distributions via spherical harmonic signals. Symphony’s performance approached that of state-of-the-art diffusion models while remaining relatively lightweight. However, it continued to face challenges in error accumulation and determining bond lengths, and it was only evaluated against small organic molecules. Here, we expand on Symphony’s capabilities and make it more compatible with larger atomic structures. We add improvements to the embedders, split the radial and angular components when predicting atom positions, and increase the radial cutoff for atomic neighborhoods considered during prediction. We also increase Symphony’s training and inference speeds through a new implementation in PyTorch, making inference nearly 4x faster than previously. In addition, we demonstrate its effectiveness across a variety of tasks, including small molecule and protein backbone generation.

Vigilis: Leveraging Language Models for Fraud Detection in Mobile Communications and Financial Transactions

Thu, 01 May 2025 00:00:00 GMT

Vigilis: Leveraging Language Models for Fraud Detection in Mobile Communications and Financial Transactions Das, Gaurab Although advances in security have strengthened defenses in digital financial systems, attackers increasingly rely on social engineering to achieve their goals. These attacks are difficult to detect and prevent with existing security measures. To address this, we propose Vigilis, a fraud-protected application that employs advanced language models to counter such attacks in calls, texts, and payments. We first collect and make available a corpus of fraudulent calls from the Internet and train lightweight transformer-based models that achieve fraud detection accuracies of up to 94% and 87% on transcript and audio modalities, respectively. We integrate these models into a real-time call system within Vigilis that operates entirely on-device, enabling accurate fraud detection in an efficient and privacy-preserving manner. We then extend Vigilis to incorporate context-aware transaction authentication, where the underlying social context behind a transaction is determined from calls, texts, and browsing history and used to infer the transaction’s validity. By uniquely incorporating social concepts into traditional cybersecurity techniques, we attempt to counter and mitigate issues related to social engineering attacks in financial fraud.

GDSVD: Scalable k-SVD via Gradient Descent

Thu, 01 May 2025 00:00:00 GMT

GDSVD: Scalable k-SVD via Gradient Descent Gan, Emily We show that a gradient-descent with a simple, universal rule for step-size selection provably finds k-SVD, i.e., the k ≥ 1 largest singular values and corresponding vectors, of any matrix, despite nonconvexity. There has been substantial progress towards this in the past few years where existing results are able to establish such guarantees for the exact-parameterized and over-parameterized settings, with choice of oracle-provided step size. But guarantees for generic setting with a step size selection that does not require oracle-provided information has remained a challenge. We overcome this challenge and establish that gradient descent with an appealingly simple adaptive step size (akin to preconditioning) and random initialization enjoys global linear convergence for generic setting. Our convergence analysis reveals that the gradient method has an attracting region, and within this attracting region, the method behaves like Heron’s method (a.k.a. the Babylonian method). Empirically, we validate the theoretical results. The emergence of a modern compute infrastructure for iterative optimization coupled with this work is likely to provide a means of solving k-SVD for very large matrices.

Comparison of dispersion metrics for estimating transcriptional noise in single-cell RNA-seq data and applications to cardiomyocyte biology

Thu, 01 May 2025 00:00:00 GMT

Comparison of dispersion metrics for estimating transcriptional noise in single-cell RNA-seq data and applications to cardiomyocyte biology Chen, Tina T. Transcription is a dynamic process with a multitude of characteristics, including transcript level, burst frequency, amplitude, and variability. Single-cell RNA sequencing data analysis often focuses on comparing transcription levels. However, these analyses capture only a portion of the wealth of information conveyed by transcription. The quantification and analysis of transcriptional variability poses an opportunity to study transcription and gene regulation from a new angle. Transcriptional variability has already been implicated in a number of biological processes, including in immune system development and in aging. Yet, the most appropriate method for measuring transcriptional variability in single-cell data has remained relatively unclear. Here, we simulated single-cell data with varying dispersion and dataset size to assess the relative responsiveness of the Gini index, variance-to-mean ratio, variance, and Shannon entropy to variability in single-cell counts. We found that the variance-to-mean ratio scales approximately linearly with increasing dispersion, and that it is scale-invariant. The Gini index displayed paradoxical behavior, and Shannon entropy was not scale-invariant. Thus, we applied the variance-to-mean to measure transcriptional variability in two publicly available datasets studying congenital heart defects in mouse models. We first found that change in transcriptional variability does not correlate with gene characteristics such as transcript level and evolutionary gene age. We also found that using change in transcriptional variability to focus GSEA and TF motif enrichment analyses revealed both genes with known involvement in cardiomyopathy and new genes and pathways as potential targets for future study. Notably, many of the genes and pathways identified through transcriptional variability analysis were not found by differential expression analysis, suggesting that transcriptional variability can provide additional biologically relevant information beyond what is observed from studying mean expression alone.

Expanding Annotation to Mixed-Media Types in a Large-Scale Social Annotation Platform

Thu, 01 May 2025 00:00:00 GMT

Expanding Annotation to Mixed-Media Types in a Large-Scale Social Annotation Platform Heiberger, Harry G. In recent years, social annotation systems have become a popular and effective tool for hosting collaborative discussions on assigned readings. One such tool created by our lab is NB. Over the last twelve years, hundreds of instructors have incorporated NB within their classes, with over 50,000 students leaving millions of annotations [1]. While feedback for NB has mostly been positive, one major limitation is its difficulty in annotating documents with nested media types. As multimodal forms of learning beyond just text are becoming increasingly common in educational assignments, having the ability to annotate beyond simple text documents would greatly increase the utility of NB in the modern classroom. This work seeks to remedy this issue by expanding the types of documents NB can successfully annotate, specifically focusing on three mixed-media issue types: independently moving text components, image annotation, and video annotation. We will explore the design space of possible implementation strategies for these features and discuss the specific design decisions that were made when adding them to NB. We hope that by increasing the types of documents NB can annotate, we will better fulfill its goal of enhancing student engagement and learning.

Pareto Task Inference Analysis of Single-Cell RNASequencing of Human Placenta Reveals Biological Insightsinto Adverse Pregnancy Outcomes

Thu, 01 May 2025 00:00:00 GMT

Pareto Task Inference Analysis of Single-Cell RNASequencing of Human Placenta Reveals Biological Insightsinto Adverse Pregnancy Outcomes Eppinger, Aria R. Adverse pregnancy outcomes (APOs), such as preeclampsia, fetal growth restriction, and preterm birth, occur in 10-15% of pregnancies. There is limited knowledge of how the cellular states in the placenta and decidua tissues are altered in women with particular APOs or may contribute to APOs. Single-cell RNA sequencing (scRNAseq) approaches have characterized cellular populations and interactions at the maternal-fetal interface using traditional dimensionality-reducing methods such as UMAP-based clustering. However, these techniques may generate limited representations of nuanced cellular functions and biological relationships among and within cell clusters. Pareto Task Inference (ParTI), a dimensionality reduction technique that fits data to an n-dimensional polygon or polytope, models how cells optimize among multiple biological functions and transition between states. We applied ParTI to assess its ability to identify nuanced cellular states and intercellular relationships and to highlight biological mechanisms underlying specific APOs. We analyzed scRNAseq data from 50 whole placental homogenates collected from healthy pregnancies and those complicated by fetal growth restriction (FGR), preterm preeclampsia (PrePET), spontaneous preterm birth (PTB), term preeclampsia or gestational hypertension (TermPET/GHTN), or type 1 diabetes (DM1). ParTI was applied to the dataset with 1) all main cell lineages (B-cells, trophoblasts, stromal, endothelial, Haufbauer, T-NK, maternal myeloid cells) and 2) syncytiotrophoblasts (SCTs), a sublineage of trophoblasts. Marker genes and gene set enrichment analysis for the ParTI polytope vertices, called archetypes, were performed to assess the biological states associated with the archetypes. We demonstrated that the ParTI polytope can separate both broad cell lineages and sublineages, suggesting that iteratively applying ParTI can serve as an alternative clustering approach when cell-lineage marker genes are previously known. Additionally, ParTI applied to SCTs separated healthy controls from pregnancies complicated by specific APOs. Gene set enrichment analysis of the cells proximal to the archetypes suggests biological differences in SCTs with specific APOs compared to the controls. Thus, ParTI can identify biological mechanisms underlying specific APOs and be applied to additional datasets to uncover biological relationships among and within cell-type clusters.

Modeling Recursion with Iteration: Enabling LLVM Loop Optimizations for Recursive Data Structure Traversal

Thu, 01 May 2025 00:00:00 GMT

Modeling Recursion with Iteration: Enabling LLVM Loop Optimizations for Recursive Data Structure Traversal Cuevas, Elie E. Recursive algorithms are a natural and expressive way to traverse complex data structures, but they often miss opportunities for optimization in modern compiler infrastructures like LLVM. This thesis explores a novel technique that temporarily transforms recursive traversals into synthetic loop-like structures, enabling existing loop-specific optimizations to apply, before transforming them back. By extending Clang’s semantic analysis and implementing a custom LLVM transformation pass, recursive traversals are initially structured into synthetic loops that can benefit from existing loop analyses and optimizations. After these optimizations are applied, the transformation restores the original recursive semantics, preserving program behavior while incorporating performance gains. Evaluation across custom microbenchmarks shows that while general recursive traversals suffer a modest overhead, workloads designed to benefit specific loop-focused optimizations achieve up to a 30% performance improvement. This demonstrates that even though the approach requires temporarily "misrepresenting" code to the compiler, selective exposure of recursive patterns to loop-based optimization infrastructure is practical and effective. This work establishes a proof-of-concept for compiler transformations that bridge recursion and iteration, paving the way for future systems that better optimize real-world recursive code without sacrificing clarity or maintainability.

Grounding Time Series in Language: Interpretable Reasoning with Large Language Models

Thu, 01 May 2025 00:00:00 GMT

Grounding Time Series in Language: Interpretable Reasoning with Large Language Models Chen, Lily Can large language models (LLMs) classify time-series data by reasoning like a domain expert—if given the right language? We propose a method that expresses statistical time-series features in natural language, enabling LLMs to perform classification with structured, interpretable reasoning. By grounding low-level signal descriptors in semantic context, our approach reframes time-series classification as a language-based reasoning task. We evaluate this method across 23 diverse univariate datasets spanning biomedical, sensor, and human activity domains. Despite requiring no fine-tuning, it achieves competitive accuracy compared to traditional and foundation model baselines. Our method also enables models to generate expert-style justifications, providing interpretable insights into their decision-making process. We present one of the first large-scale analyses of LLM reasoning over statistical time-series features, examining calibration, explanation structure, and reasoning behavior. This work highlights the potential of language native interfaces for interpretable and trustworthy time-series classification.

National crop field delineation for the United States

Thu, 01 May 2025 00:00:00 GMT

National crop field delineation for the United States Chen, Zitong Comprehensive and accurate crop field boundary maps are crucial for digital agriculture, land management, and environmental monitoring. However, no high-quality field boundary dataset is publicly available in the United States. This thesis addresses this gap by creating a new, large dataset and training a deep learning model capable of mapping field boundaries. We built a dataset of over 15,000 image-mask pairs using high-resolution National Agriculture Imagery Program (NAIP) satellite imagery and curated field boundary labels. This dataset covers a variety of leading agricultural states and includes images taken at different scales to capture a wide variety of field sizes and layouts. We used this dataset to train an adapted ResUNet++ neural network model designed to segment crop fields. The trained model achieved around 0.8 for pixel-level accuracy, showing it can generally identify field areas well. However, its performance in matching predicted individual field instances with the ground truth instances (measured by mean instance Intersection over Union, or mIoU) was around 0.5. This lower instance score was largely due to the post-processing step, which converts the model’s probability predictions into separate field instances. Despite this, the field polygons produced by our approach are visually coherent with satellite field images and can be readily used with geospatial tools like Google Earth Engine. Our work provides a practical starting point for future research on mapping fields across the contiguous U.S. Potential directions for improvements may involve developing sharper boundary predictions, exploring direct instance segmentation models, refining post-processing methods, and expanding the dataset to include more challenging areas.

Design and Implementation of an Analog High Power Broadband Self Interference Canceller for In Band Full Duplex

Thu, 01 May 2025 00:00:00 GMT

Design and Implementation of an Analog High Power Broadband Self Interference Canceller for In Band Full Duplex Hanly, Bianca Marie A Self-Interference Canceler is the principle component that allows for Simultaneous Transmit And Receive (STAR) of radio signal broadcasting. Previous research and designs by other groups have resulted in systems that either operate at high powers or are capable of cancellation over a wide bandwidth. This work seeks to build upon previous research in order to design an analog SIC that is capable of both high power (∼100W) and wide instantaneous bandwidth (∼1GHz) cancellation. The system is designed as a vector modulator using off-the-shelf hybrid couplers and switches with a custom variable attenuator designed using PIN diodes in a Waugh attenuator architecture. The system was fabricated on a four layer PCB and measured with a network analyzer. Simulated results for variable attenuator and overall vector modulator are presented.

Implementation of Semantic SLAM on a Mobile Manipulator System

Thu, 01 May 2025 00:00:00 GMT

Implementation of Semantic SLAM on a Mobile Manipulator System Francis, Zachary R. In the field of robotics, the development of household robots capable of performing everyday tasks continues to be a major area of research and practical interest. Many domestic chores—such as picking up and moving objects from one location to another—have been successfully performed by stationary robotic manipulators paired with visual perception systems. However, accomplishing more complex, varied, and spatially distributed tasks in real-world home environments requires a mobile platform with a more human-like form factor. These tasks demand greater flexibility, spatial awareness, and interaction capabilities than fixed systems can typically provide. This work focuses on the RBY1 robot from Rainbow Robotics, a humanoid platform designed to support advanced manipulation and mobility. A range of tools and modules were developed to enhance its functionality, including software for semantic perception, task execution, and environment interaction. This thesis provides a technical overview of these tools, highlighting their roles in collecting new datasets that can be used for semantic SLAM research. In the future, these tools can enable the robot to operate more effectively in domestic settings, towards the ultimate goal of enabling more capable home-assistive robots.

Improving Introductory Computer Science Students’ Programming Process When Using a Generative AI Tutor (PyTutor)

Thu, 01 May 2025 00:00:00 GMT

Improving Introductory Computer Science Students’ Programming Process When Using a Generative AI Tutor (PyTutor) Cunningham, Caroline K. This thesis examined students’ programming process while using PyTutor, a generative AI tutor for introductory computer science students. This thesis had the research questions: (1) How does the process of test case creation, with or without PyTutor’s Test Case Runner, impact students’ programming process while using PyTutor? (2) How can prompt engineering of PyTutor’s system prompt be leveraged to improve AI Chat response quality with respect to: (a) reducing the amount of code revealed in the answer, (b) improving the conciseness of responses, and (c) having the AI chat give the student test cases as a tool to understand code correctness? (3) How does PyTutor’s responses from the updated prompt affect the programming process for computer science students? A key finding from a focus group in the first stage (n=9) was apart from test cases and was that the majority of participants who asked questions to PyTutor received at least three lines of code, unideal for PyTutor’s pedagogical purpose. This discovery inspired the next phase of this thesis of prompt engineering PyTutor, which resulted in an updated prompt. Responses from the both the updated prompt and the original prompt were scored using an evaluation rubric. For the Students thinking through problem category of the evaluation rubric, it was statistically significant that the distribution of points for responses from the updated prompt was greater than the distribution of points for responses from the original prompt. Finally, participants were asked to solve a programming problem using either PyTutor with the updated prompt (n=10) or PyTutor with the original prompt (n=2). Across the focus groups from the first and final stage, I found that fewer participants who used PyTutor with the updated prompt received at least three lines of code. Furthermore, participants who used PyTutor with the updated prompt required a greater number of messages to first receive three lines of code. Additionally, all four participants who received at least three lines of code from PyTutor with the updated prompt asked majority high-level questions. As participant feedback suggested that PyTutor’s responses for high-level questions could be repetitive, this data highlights a new direction of improving PyTutor’s responses when answering high-level questions to benefit students’ programming process.

Initial segments in ordinal recursion theory.

Mon, 01 Jan 1979 00:00:00 GMT

Initial segments in ordinal recursion theory. Dorer, David John. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Mathematics, 1979; Vita.; Bibliography: leaf 49.

Tool and chip temperatures in machine shop practice

Tue, 01 Jan 1924 00:00:00 GMT

Tool and chip temperatures in machine shop practice Shore, Henry. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1924; Includes bibliographical references.

The oxidation of sulphur dioxide in Cottrell precipitators of a contact sulphuric acid plant

Thu, 01 Jan 1931 00:00:00 GMT

The oxidation of sulphur dioxide in Cottrell precipitators of a contact sulphuric acid plant Haberstoh, Robert H.; Milligan, Sydney.; Roever, Paul H. Thesis: B.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1931

Production of carbon black by the decomposition of methane with electrically heated wires

Sun, 01 Jan 1939 00:00:00 GMT

Production of carbon black by the decomposition of methane with electrically heated wires Donatello, Dominic G. Thesis: B.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1939; Includes bibliographical references (leaf 24).

Models for investigating the unreliability of freight shipments by rail.

Sat, 01 Jan 1972 00:00:00 GMT

Models for investigating the unreliability of freight shipments by rail. Folk, Joseph Frederick. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Electrical Engineering, 1972; Vita.; Bibliography: leaves 279-284.

A set-theoretic approach to state estimation.

Wed, 01 Jan 1969 00:00:00 GMT

A set-theoretic approach to state estimation. Hnyilicza, Esteban. Thesis: M.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1969; Bibliography: leaves 112-113.

Design of a voltage-limiting device using SIC nonlinear resistors.

Tue, 01 Jan 1974 00:00:00 GMT

Design of a voltage-limiting device using SIC nonlinear resistors. Asamoah, William Kafui. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1974; Includes bibliographical references.

An analog circuit simulator for the Connection Machine

Thu, 01 Jan 1987 00:00:00 GMT

An analog circuit simulator for the Connection Machine De Beus, Eric. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 1987; Includes bibliographical references.

Boiling heat transfer in rotating channels with reference to gas turbine blade cooling

Sun, 01 Jan 1984 00:00:00 GMT

Boiling heat transfer in rotating channels with reference to gas turbine blade cooling Mudawar, Issam Abdallah. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1984; Includes bibliographical references.

A study of the developments in the construction, equipment and operation of street railway cars

Thu, 01 Jan 1920 00:00:00 GMT

A study of the developments in the construction, equipment and operation of street railway cars French, Grant Keith. Thesis: B.S., Massachusetts Institute of Technology, Department of Civil Engineering, 1920

High Precision Binary Trait Association on PhylogeneticTrees

Thu, 01 May 2025 00:00:00 GMT

High Precision Binary Trait Association on PhylogeneticTrees Balogun, Ishaq O. Understanding how genetic variation drives microbial phenotypes is fundamental to advancing microbiology, particularly in pathogenicity, drug resistance, and host adaptation. Traditional genome-wide association study (GWAS) methods fail to account for shared evolutionary history, confounding association analyses. Microbial GWAS approaches emerged to address this, but modern methods often lack the statistical power to detect associations while controlling false discoveries, and face computational limits at scale. Here, we present SimPhyNI (Simulation-based Phylogenetic iNteraction Inference), a computational framework for detecting binary trait-trait associations in microbial populations. SimPhyNI uses stochastic simulations of trait evolution on phylogenetic trees to detect positive and negative associations with high precision and recall. Benchmarking on large synthetic datasets, SimPhyNI achieved a precision-recall AUC (PR AUC) of 0.987 and 0.975 for positive and negative interactions, respectively, indicating near-perfect discrimination of true from neutral associations. Competing methods showed substantially lower performance, especially for negative associations. We further applied SimPhyNI to empirical datasets, recovering known biology and generating plausible hypotheses for novel mechanisms. Though tested here on binary traits, SimPhyNI’s design supports future extension to multi-state and continuous traits using generalized models. Its high recall also makes it well-suited for constructing gene interaction networks and identifying co-evolving trait modules. By combining evolutionary modeling with scalable statistics, SimPhyNI advances our ability to uncover the genetic interactions that drive microbial function, ecology, and disease.

Foundations for Building an Innovation-Centric Product Development Framework for Medical Devices

Thu, 01 May 2025 00:00:00 GMT

Foundations for Building an Innovation-Centric Product Development Framework for Medical Devices Rajan, Neena E. The medical device industry, governed by a tight regulatory landscape, often relies heavily on structured Product Development Processes (PDPs) to bring innovative solutions to market. These structured processes create significant challenges when integrating technological innovations that emerge in the later stages of the development cycle. This study explores the complexities of this "innovation paradox" within large United States-based medical device corporations, examining how the rigidity of traditional PDP models affects the incorporation of innovative changes to in-flight projects. Drawing upon insights from a comprehensive literature review and a quantitative analysis utilizing a Monte Carlo simulation, this research highlights the impact of integrating an innovative change on the overall project timeline and cost. The simulation results show that introducing innovative changes to the PDP typically extends project timelines and increases the total net present costs and are affected by the timing of the change and its technological maturity. Introducing changes in later project phases significantly increases both duration and cost compared to earlier phases. Changes with lower technological maturity led to greater duration and cost escalations, especially when introduced late in the development cycle. To balance regulatory requirements and PDP agility, large medical device companies can adopt hybrid PDP models, establish dedicated innovation assessment teams, create flexible product designs, and focus on value-driven innovations that meet patient and market needs.

Savaal: A system for automatically generating high-quality questions from unseen documents

Thu, 01 May 2025 00:00:00 GMT

Savaal: A system for automatically generating high-quality questions from unseen documents Chandler, Joseph A. Assessing human understanding through exams and quizzes is fundamental to learning and advancement in both educational and professional settings. However, current solutions to automate the generation of challenging questions from educational materials and documents are insufficient, resulting in superficial or often irrelevant questions. While LLMs have been shown to excel in tasks like question answering, their usage on question generation is underexplored for general domains and at scale. This work presents Savaal, a scalable question-generation system that generates higher-order questions from documents, as well as a real-world system implementation for general use. Savaal accomplishes the following goals and objectives: (i) scalability, capable of generating hundreds of questions from any document (ii) depth of understanding, synthesizing higherorder concepts to test learners’ understanding of the material, and (iii) domain independence, generalizing broadly to any field. Rather than naively providing the entire document in context to an LLM, Savaal breaks down the process of generating questions into a three-stage pipeline. We demonstrate that Savaal outperforms the direct prompting baseline as evaluated by 76 human experts on 71 documents across conference papers and PhD dissertations. We additionally contribute a general system for serving Savaal in real-world scenarios. We demonstrate that our system is scalable, enabling fault-tolerant and horizontal scaling of each individual component in response to fluctuations in usage. Moreover, our architecture enables interactive usage from users and collaboration in groups, reflecting real-world organizations like classrooms or enterprises. We hope that the system enables scalable question generation for educational and corporate use-cases.

Multi-Objective Exploration of Refueling Architecture for Sustainable Crewed and Cargo Space Transportation

Thu, 01 May 2025 00:00:00 GMT

Multi-Objective Exploration of Refueling Architecture for Sustainable Crewed and Cargo Space Transportation Terakado, Daiki This thesis presents a new integrated framework for evaluating in-space refueling architectures, focusing on their application to the human space missions such as Artemis. The framework tightly couples vehicle sizing with a boil-off control model, allowing the evaluation of various combinations of propellant types, refueling locations, and boil-off control. The model captures the dynamic interdependence between the components of the refueling system, the transport vehicle, the refueler, and the depot, using an iterative approach to ensure consistent mass estimates across configurations. The framework is applied to analyze human landing system (HLS) architectures with refueling in cis-lunar space. The key findings highlight the mass savings benefits of cryocoolers, the benefits of high Isp with Lox/LH2, the benefits with NRHO refueling for acceptable ΔV requirement, and positive and negative effects of reusability in mass and mission time. Furthermore, the study indicates that the number of required refueling events is more sensitive to payload and refueler capacity than to boil-off losses. To extend the framework toward long-term, scalable transportation solutions, the thesis compiles a comprehensive set of figures of merits (FoMs) and discusses future model extensions including risks, ISRU, and electric propulsion. Limitations such as lack of reusable configuration flexibility, and insufficient support for Mars mission parameters are identified as areas for future development. This work provides a foundational framework for the exploration of refueling architecture and solid next steps to design sustainable and scalable human space transportation systems.

Embedded Software-Defined Radio Architectures for 6G Cellular Networks

Thu, 01 May 2025 00:00:00 GMT

Embedded Software-Defined Radio Architectures for 6G Cellular Networks Urbonas, Jonas Over the past decades, the widespread adoption of wireless communication technologies in the industrial, scientific, medical, defense, and commercial sectors has resulted in substantial advancements in digital radio technologies. Each new generation of cellular technology, beginning with 1G, has introduced novel use-case scenarios that have challenged the performance of the prevailing digital radio architectures. The newly proposed scenarios for 5G-Advanced, and the upcoming 6G cellular networks due to be standardized by 2030 are no exception. The emerging 6G network components, such as the space-air-ground integrated cell-less networks, as well as the artificial intelligence-native network architecture, drive the demand for flexible and fully reconfigurable radio units supporting multi-GHz instantaneous signal bandwidths, frequency agile radio architectures covering multi-octave frequency ranges, and highly sensitive receivers. To support these requirements, software-defined radios (SDR) are becoming an essential building block of next-generation radio networks. This thesis presents a review of softwaredefined radio technology, examines its history, proposes the requirements of SDR units for 6G cellular networks, and presents a quantitative performance analysis of over 2 million distinct SDR architectures that could be used in 6G communication networks. It does so by defining the key system architectural decisions and their options, including the data converters, filters, mixer and amplifier technologies. It also examines different radio transmitter and receiver architectural topologies, including baseband sampling, IF sampling, direct RF sampling, and fully digital RFSoC, and constructs a multi-attribute utility (MAU) to quantify the system performance. The MAU is used to build a tradespace of SDR architectures, enabling the identification of the Pareto frontier. Analysis of SDR system architectures on the Pareto frontier reveals that the performance of direct RF sampling SDR architectures is highly competitive with industry-standard IF sampling. The tradespace is also used to analyze the sensitivity of system performance to individual architectural decisions via a main-effect analysis, allowing quantification of connectivity and sensitivity of available architectural decisions. Sensitivity analysis reveals that system performance is highly sensitive to receiver architectural decisions, particularly analog-to-digital converters, indicating the need for continued advances in this technology to produce high-performance SDR systems.

Computational Design of Architected Lattices for Construction Applications

Thu, 01 May 2025 00:00:00 GMT

Computational Design of Architected Lattices for Construction Applications Leamon, Sophie Architected lattices have been utilized in aerospace and research applications for their modularity, scability, reconfigurability, and high strength-to-weight properties. However, voxels have yet to find widespread integration in the residential or commercial construction industry because of the industry’s distinct system needs. This study identifies the pain points unique to the construction industry that have slowed or disabled the adoption of new practices, highlighting the importance of utilizing known materials, methods, and the transparency of the design process, as major hurdles to adoption of innovation in the industry. This study presents a computational approach to designing architected lattices that seeks to undermine these core issues by making building with architected lattice structures agnostic to material and manufacturing methodology. Three open source computational approaches to architectural design are proposed: 1) integration of support structures for additively manufactured structures; 2) parametric design of voxels from 2D material, their manufacturing molds, and optional alignment features; and 3) generation of two-dimensional cut files for assembly with 3D printable joinery. These files are computationally designed and arranged for instantaneous production to demystify the lattice architectural design process, establish a pathway for utilizing all available materials in lattice construction, reduce the overhead costs for experimentation with lattice structures, and eliminate barriers to the fabrication process by enabling accessible manufacturing methods.

Evaluating the Strategic Intent and Competitive Dynamics of China’s Satellite Communications Constellations

Thu, 01 May 2025 00:00:00 GMT

Evaluating the Strategic Intent and Competitive Dynamics of China’s Satellite Communications Constellations Delkowski, Michal . This thesis examines the strategic, technical, and economic feasibility of China’s two flagship low Earth orbit (LEO) satellite megaconstellation programs, Guowang and Qianfan, in the context of the rapidly evolving global satellite communication (Satcom) market. Against the backdrop of SpaceX’s Starlink dominance and intensifying geopolitical competition, China’s efforts represent not only a telecommunications infrastructure push but also a broader assertion of technological sovereignty and global influence. This study uses a scenario-based analysis that integrates system throughput analysis and financial forecasting. Three deployment scenarios (base, optimistic, and pessimistic) are analyzed, accounting for satellite production rates, launch capabilities, and regional adoption patterns, particularly across Belt and Road Initiative (BRI) markets. The study also evaluates "system-of-systems" integration with China’s military objectives, and spectrum coordination challenges. Key findings reveal that Guowang becomes marginally viable only in the optimistic scenario, assuming deployment of at least 9,000 satellites, reduced satellite unit costs (targeting ~$300,000 per satellite), expanded gateway infrastructure, and realization of these targets by 2035, while remaining unviable in base and pessimistic cases. Qianfan faces greater commercial risk, achieving viability only with early adoption in BRI countries and government dual-use contracts, incurring a pessimistic-case NPV loss exceeding $76B. Resource allocation problem (RAP) modeling suggests that projected throughput may saturate early without major gateway expansion. Both constellations require China to scale reusable rockets and sustain a combined annual launch rate exceeding 1,000 satellites by the early 2030s. Neither constellation system meets China’s 2030 rural broadband targets under base-case conditions, over 40% of the 336M unconnected citizens remain underserved without terminal subsidies. Ultimately, China’s LEO Satcom strategy depends not on satellite count alone but on coordinated progress in launch economics, affordability, dual-use policy, and international partnerships.

Optimization of CPG budgets in Retailer-led marketing programs

Thu, 01 May 2025 00:00:00 GMT

Optimization of CPG budgets in Retailer-led marketing programs Gandhi, Abhinav Grocery retailers and Consumer Packaged Goods (CPG) companies have a symbiotic relationship. Retailers need CPGs to supply the products, and CPGs need retailers’ customers to grow their brands. Since shelf space is limited, CPGs offer trade and marketing funds to prominently feature their brands. As part of loyalty programs, retailers offer coupons to customers that are often funded by CPGs. In return, CPGs expect a return on their investment(ROI). Since budgets are limited and are also expected to be utilized, it becomes a challenge for the retailer to find the right size of a mailer which can balance costs and relevance to customers. This thesis explores how knapsack problems can be used in an non-adaptive setting to help maximize the reach of print and email campaigns. Seeking inspiration from existing literature, multiple simulations were set up to evaluate budget-constrained allocation and compare two approaches, the multiple-choice Knapsack (MCK) and a greedy algorithm. Considering uncertainty in redemption, the Newsvendor model was also explored to review the possibility of over-allocation to improve budget utilization and increase reach. The preliminary analysis findings offer promising results and provide a setting for further research.

Exploration of design strategies and optimization for efficient mass timber structures as a function of column position

Thu, 01 May 2025 00:00:00 GMT

Exploration of design strategies and optimization for efficient mass timber structures as a function of column position Gerken, Christoph The building sector is responsible for a large share in global carbon emissions. As the load bearing structure is particularly material-intensive, a decisive shift can be achieved by improving its design and decreasing its volume. This thesis examines how structural mass timber floor systems can be designed in an efficient, low-waste manner through a design-oriented approach that is immediately applicable within the context of conventional construction techniques and building practices. Reducing material in timber structures has economical and ecological benefits. Reduced timber demand entails significant cost savings and decreased building weight which considerably cuts embodied carbon. Since common floor systems mainly act in bending, this work focuses on the reduction of moment forces in standard setups comprised of timber slabs, beams, and columns. In principle, bending forces in beams and slabs can be reduced by moving the supports inwards, leading to overhanging structural elements. The original method presented in this thesis shows how this approach applies to conventional mass timber floor systems. This work provides an understanding of how informed column positioning can take advantage of this behavior and allows for material and embodied carbon reduction trough design. The consequent architectural implications of the resulting irregular column grid are explored in a floor plan design suggestion Material demand and embodied carbon are evaluated as a function of column position through finite element analysis and optimization as part of a computational model. By consulting a mass timber manufacturer’s catalogue to assign appropriate products to structural members, this approach enables material reduction in the design process rather than in the production. Bypassing slow-changing, inert fabrication procedures, this method can be realized instantaneously. This work identifies the optimal support position to reduce bending forces in beams and slabs to be at 41% of the distance from the element’s edge to its midspan. Furthermore, this research finds that the impact of ideal column position on material efficiency depends on required minimum effective spans. While being negligible in the absence of constraints, informed column positioning can reduce timber demand by 20% and embodied carbon by 16% when subjected to a minimum effective span requirement of 6 m – a common span in timber construction – in a building of 30x30 m and five floors. Building dimensions are found to have an insignificant impact on these results. This thesis illustrates the potential for architects and engineers to enhance structural efficiency of mass timber floor systems merely by deviating from the usual, regular column grid and taking advantage of straightforward structural principles through design.

Machine Learning for the Condition Assessment of Concrete Bridges

Thu, 01 May 2025 00:00:00 GMT

Machine Learning for the Condition Assessment of Concrete Bridges Fayad, Fred The assessment of concrete bridge conditions is critical for ensuring structural integrity and public safety. Traditional inspection methods, which rely heavily on visual inspections and manual assessments, are time-consuming, subjective, and prone to human error. With the increasing number of aging bridges worldwide, there is a growing need for more efficient and accurate methods to assess bridge health. This thesis aims to explore the application of machine learning techniques for automating the bridge condition assessment process and improving the accuracy and reliability of bridge evaluations. This study investigates the development and implementation of a model consisting of two machine learning algorithms to predict the condition of concrete bridges based on data collected from various public sources. The first algorithm appraises the structural health of a bridge based on bridge rating and the second algorithm assesses the condition of a bridge after a specific failure mechanism. Specifically, this work focuses on using classification algorithms such as Random Forest (RF), XGBoost, and Neural Networks (NN) in both algorithms to achieve their purpose. The results of this study demonstrate that machine learning models can provide a decent performance in predicting bridge conditions. The overall model achieved a testing accuracy of 79%. This research contributes to the field of civil engineering by showcasing the potential of machine learning in infrastructure management. By automating the assessment process, the proposed models can help reduce the time and cost of inspections while providing more accurate data to guide maintenance planning and bridge rehabilitation efforts. Future work will focus on further optimizing the models, incorporating additional data sources, and deploying the system for real-time bridge monitoring.

A Comparison of Theoretical and Actual Coumarin Exudation Under Iron Limitation to Understand Root Exudation Mechanics

Thu, 01 May 2025 00:00:00 GMT

A Comparison of Theoretical and Actual Coumarin Exudation Under Iron Limitation to Understand Root Exudation Mechanics Van Note, Lana Nutrient cycling is an important component of plants’ immune systems, largely driven by the act of exuding environmentally influential metabolites from roots. Root exudation may be driven by multiple unique mass-transport mechanisms, including active and passive transport types, though the latter is not well-studied despite being labelled a significant driver of low molecular weight metabolite exudation. This research investigates the generally accepted assumption that low molecular weight metabolites, including iron-fixing coumarins (scopoletin, fraxetin, etc.) are primarily exuded passively, and high molecular weight metabolites follow an active exudation approach. Scopoletin and scopolin exudation from Arabidopsis thaliana in low-iron and replete conditions is quantified to determine if the hypothesized passive diffusion mechanism is a significant contributor to coumarin exudation. LC-MS analysis suggests that passive diffusion of scopoletin and scopolin from roots plays a significant role in total coumarin exudation values. Further research should include investigating the implications of passive coumarin exudation on long-term iron storage and soil health in addition to the relationship between coumarin production and exudation.

Offshore floating solar with compressed air storage as a baseload power plant for a data center

Thu, 01 May 2025 00:00:00 GMT

Offshore floating solar with compressed air storage as a baseload power plant for a data center Athanasopoulos, Panagiotis Rafail This thesis presents the conceptual design, technical modeling, and economic analysis of a novel offshore floating solar energy system integrated with Compressed Air Energy Storage (CAES) for reliable baseload power delivery to coastal data centers. The system architecture is modular, consisting of multiple “powercells,” each comprising a 5×5 photovoltaic (PV) array mounted above a matrix of submerged compressed air storage cylinders anchored below the floating platform, addressing the energy resilience and spatial constraints of coastal computing infrastructure. This scalable configuration enables distributed energy collection and localized storage, tailored to meet site-specific demands. Detailed thermodynamic modeling of both charging and discharging cycles is conducted, with analytical solutions validated against a full numerical implementation. Results show that under realistic operating assumptions, the temperature inside the storage vessels remains nearly isothermal due to the long charging duration and large heat exchange surface, enabling a simplified energy balance model. A techno-economic analysis evaluates both structural steel requirements and photovoltaic investment, benchmarked against market data from 2024. Key metrics such as structural cost per unit energy ($/kWh) and per rated power output ($/kW) are derived. The hybrid system is found to be economically competitive with lithium-ion (Li-ion) battery alternatives, offering extended lifespan (20–30 years), lower material costs, and enhanced sustainability through avoidance of critical minerals. Environmental and mooring considerations for offshore deployment are also addressed, demonstrating the feasibility of integrating energy generation, storage, and maritime infrastructure. This work advances the development of resilient, decarbonized energy systems aligned with global renewable energy targets and the rising demand for sustainable data center operations.

Destructive Behaviors in Naval Shipyards: A STAMP and System Dynamics Analysis

Thu, 01 May 2025 00:00:00 GMT

Destructive Behaviors in Naval Shipyards: A STAMP and System Dynamics Analysis Brower, Braden C. United States Navy Refueling and Complex Overhauls (RCOHs), and other extended maintenance availabilities, present uniquely demanding environments where Sailors face elevated risks for destructive behaviors, including suicide and substance abuse. Prolonged exposure to harsh industrial conditions, significantly degraded Quality of Service, demanding workloads, and critical manning shortfalls create cumulative stress distinct from operational duty. These destructive behaviors severely impact personnel’s well-being, erode force readiness through attrition and morale issues, and indicate systemic contributing factors as highlighted by recent investigations into carrier suicides during shipyard periods. This thesis utilizes Causal Analysis based on Systems Theory (CAST), grounded in systems thinking, to analyze the USS George Washington RCOH events and identify the underlying safety control structure flaws that contributed to this hazardous environment. Insights from the CAST analysis were then integrated with a qualitative System Dynamics model to better understand the feedback loops and dynamic interactions driving system behavior, particularly revealing a capability trap dynamic exacerbated by resource constraints and personnel pressures. The analysis identified critical, interacting systemic flaws across multiple organizational levels that contributed to the accident: (a) inadequate strategic resourcing and manning prioritization for RCOH personnel support, (b) deficient planning, risk management, and oversight processes that were ineffective at protecting Sailor well-being amidst budget and schedule pressures, (c) ineffective feedback mechanisms that prevented critical information from reaching decision-makers, (d) and reliance on flawed assumptions regarding the RCOH environment, Sailor resilience, and standard process adequacy. Based on these findings, the thesis provides actionable, systemically focused recommendations aimed at strengthening the Navy's safety control structure by improving decision makers’ mental models, enhancing feedback and oversight, enforcing well-being constraints, and fostering organizational learning. Combined, these recommendations empower leaders to proactively manage risks, reduce destructive behaviors, and ensure a safer, more resilient environment during future RCOHs.

Enhancing Community Risk Preparedness for Flooding Emergencies: A System Dynamics Approach for the U.S. Army Corps of Engineers

Thu, 01 May 2025 00:00:00 GMT

Enhancing Community Risk Preparedness for Flooding Emergencies: A System Dynamics Approach for the U.S. Army Corps of Engineers Hoyt, Thomas S. Flooding events pose a significant and growing threat to communities in the United States, particularly as climate change alters weather patterns and sea levels continue to rise. This thesis examines how the U.S. Army Corps of Engineers (USACE) can enhance community preparedness for flood emergencies through improved risk communication strategies. Focusing on the New England District as a representative case, it integrates data from the Federal Emergency Management Agency’s (FEMA) National Household Survey and the National Flood Insurance Program (NFIP) claims archive to develop and calibrate a System Dynamics model of flood risk perception and preparedness. The model built in this thesis incorporates key variables and captures the feedback loops that influence community preparedness over time. Scenario testing demonstrates that monthly to quarterly engagements by USACE help sustain risk awareness and reduce flood-related damage, whereas less frequent engagement demonstrates minimal improvement above the baseline. By contrast, barriers to action, such as complex procedures or limited access to information, can substantially slow the adoption of preparedness measures. High levels of trust in authorities further amplify the effectiveness of risk communication and foster community engagement. This model quantifies the importance of frequent engagement, low barriers to action, and trust-building initiatives in reducing flood impact. Through calibration against historical claims and survey data, the model provides a robust framework that can guide USACE and partner agencies in refining their own flood risk communication strategies to bolster community resilience.

Structural engineering model of irregular and efficient concrete beams: Application to topology optimized shapes and integrated textile reinforcement

Thu, 01 May 2025 00:00:00 GMT

Structural engineering model of irregular and efficient concrete beams: Application to topology optimized shapes and integrated textile reinforcement Stribos, Sophia Concrete remains one of the most widely used construction materials due to its strength, durability, and availability. However, it is responsible for a large share of the global carbon emissions. Within the 40% of the global emissions attributed to the building sector, 5-8% alone accounts for the production of cement, a key component in concrete. As the construction industry seeks innovations towards sustainable practices, alternative beam designs that improve material efficiency and introduce nontraditional reinforcement systems are emerging as promising potential. However, accurate structural models capable of predicting and validating the performance of these innovative beams are often lacking, limiting their implementation in the industry, primarily due to safety and code compliance. This thesis bridges this gap by developing and validating a structural engineering model to predict the shear and flexural capacities and the deflection of irregular, efficiently shaped concrete beams, including those with alternative reinforcement and formwork. The model discretizes a 3D beam geometry into 2D sections to perform a geometric and structural cross-sectional analysis along the beam’s length. The structural engineering model is applied to two case studies: a topology-optimized steel-reinforced concrete beam and an integrated knit textile reinforced concrete beam, using experimentally measured material properties and beam testing data. The predicted engineering model results are compared against experimental data to validate the model’s accuracy. While the model could accurately capture the behavior of the topology-optimized steel-reinforced beam, it slightly overestimated the strength of the knit-textile reinforced beam. The engineering model for the topology-optimized beam had a close alignment in flexural capacity and had a slightly conservative estimate in shear and deflection due to the nature of the design equations. However, the model showed a minor overprediction in the flexural capacity and deflection of the integrated knit textile beam. Discrepancies in this model were linked to inaccurate material properties, experimental imperfections, and prestressing effects. To ensure complete accuracy and reliability, additional beam analysis using this model is needed. This research advances structural design by offering a tool for predicting the capacity and serviceability of irregular, efficiently shaped concrete beams, including those with alternative reinforcement. This thesis enables designers to validate and optimize their innovative beam designs and support their ideas as sustainable solutions within the concrete construction industry.

Assessment of Decarbonization Pathways of Japan

Thu, 01 May 2025 00:00:00 GMT

Assessment of Decarbonization Pathways of Japan Suto, Sadami Developing realistic pathways for decarbonization is crucial for the success of climate change mitigation actions. To evaluate Japan’s pathways toward achieving carbon neutrality, this study enhances the MIT Economic Projection and Policy Analysis (EPPA) model and analyzes a suite of policy scenarios that combine domestic mitigation measures such as emissions targets from the updated Japan’s Nationally Determined Contribution (NDC), power mix goals, and availability of carbon capture and storage (CCS) with international emissions trading. The impacts on CO₂ emissions, GDP, consumption, carbon prices, and sectoral output in Japan between 2030 and 2050 are assessed. Under the baseline scenario, emissions over time remain flat at about 1,000 MtCO₂e, far exceeding the carbon neutrality goal. Even when Japan’s 2030 and 2040 NDC for CO₂ and power mix targets are fully achieved, residual emissions of 100 – 200 MtCO₂e remain, which calls for a need of carbon offsets. Relying on domestic-only measures is costly for Japan. In high-ambition domestic-only scenarios without CCS, carbon prices soar to over $46,000/tCO₂ by 2050, leading to GDP losses exceeding $1.5 trillion (23% of GDP) and significant contractions in key sectors of the economy. In contrast, scenarios incorporating international emissions trading enable Japan to achieve comparable total emissions reductions by partially relying on imported carbon credits. This mechanism significantly lowers marginal abatement costs, allowing carbon prices to stabilize at $20 –$30/tCO₂ and reducing GDP losses to about $100 billion (1.6% of GDP) by 2050. Scenarios that emphasize domestic reductions while flexibly using international credits emerge as manageable pathways. These scenarios achieve domestic emissions reductions of 40 – 60% by 2050, with carbon prices ranging from $140 to $340/tCO₂ and GDP losses contained between $150 and $290 billion (2.3% and 4.3% of GDP). Importantly, these scenarios incorporate the deployment of CCS, which plays a critical role in reducing marginal costs and enabling deeper abatement in hard-to-decarbonize sectors. Most industrial sectors maintain stable output, while carbon-intensive sectors undergo gradual structural transitions. Overall, these findings suggest that Japan can achieve carbon neutrality through an integrated strategy that combines strengthened domestic action, technological deployment, and international cooperation. This study provides a robust quantitative foundation for designing feasible, equitable, and cost-effective climate policies.

Impact-Induced Bridge Failures: Analyzing Structural Vulnerabilities and Optimizing Pier Designs for Enhanced Resilience

Thu, 01 May 2025 00:00:00 GMT

Impact-Induced Bridge Failures: Analyzing Structural Vulnerabilities and Optimizing Pier Designs for Enhanced Resilience Ren, Daisy Due to the rise in global traffic patterns in recent years, bridge failures due to impact effects are becoming an increasing concern, especially for aging infrastructure. Following the recent collapse of the Francis Scott Key Bridge, issues regarding bridge vulnerabilities and design deficiencies arose, which highlighted the need for better design codes and protection for bridge piers. This study aims to address these issues by better understanding bridges' impact-related structural failure mechanisms by developing a comprehensive optimization framework to enhance the resilience of structures to dynamic impact forces using three phases: (i) statistical analysis of bridge failure data from the Multidisciplinary Center for Earthquake Engineering Research (MCEER), with data focusing on the frequency, bridge types, and bridge material trends associated with different bridge failures across the United States, (ii) development of a compliance-based optimization for trusses using MATLAB that is applied to 2D representations of pier structures for different truss configurations (2X3, 3X4, 3X5) under stress, load, and volume constraints to simulate large magnitude impact conditions, and (iii) design and validation of optimization results through mathematical calculations of compliance and strain energy to ensure consistency between numerical results and structural mechanics principles. Both fail-safe and shape optimization strategies are employed and compared across all truss configurations, revealing distinct design methodologies between maximum and minimum compliance optimizations and the trade-offs between stiffness and energy dissipation. Maximum compliance optimization designs demonstrate increased redundancy and strain energy capacity, while minimum compliance optimization designs showed increased efficiency but were more prone to brittle failure. The final study utilizing volume constraints further examined material distribution under realistic impact loads and highlighted the importance of distributed load paths and deformation capacity in structural performance. This work provides a design framework for energy-absorbing pier geometries and aims to offer insight into improving current design standards for pier designs to account for extreme events and help guide retrofitting efforts that could prevent future failures.

Computing Economic Equilibria and Their Applications to Market Games

Thu, 01 May 2025 00:00:00 GMT

Computing Economic Equilibria and Their Applications to Market Games Bruce, Samuel G. The emergence of new technologies such as e-payments and tokenized assets, distributed ledgers, smart contracts and encryption have created new opportunities for improving access and equity in financial institutions. These new tools can be used to build better infrastructure and improve economic efficiency, especially in previously underdeveloped countries. The use of these tools in various applications however requires and intimate link between economics and computer science to ensure an implementation that is both computationally efficient and improves social welfare. There has been significant research in the field of computer science concerning the computation of economic equilibria, specifically Nash Equilibria and Correlated Equilibria. These algorithms, however, have not been used in many financial applications. Further, while research exists on various methods of computation for Correlated Equilibria, little exploration has been done evaluating the quality of these equilibria in terms of economic efficiency in specific mechanisms. This work provides a sweeping view of the existing literature on equilibrium computation as well as an analysis on the economic and algorithmic tradeoffs of different approaches. The discussion begins with simple 2-player, finite action games, then moves to more complex machine learning based method for equilibrium computation in difficult settings. One of these methods is then extended to a limit-order market game explicitly described by Dubey [1] and implemented, with small modifications, by SPEEDEX [2]. This limit-order game offers a continuous, vector-valued action space with complex payoff functions, causing tension with many of the equilibrium computation algorithms explored previously. This paper identifies these tensions, then offers modifications to algorithms which allow tractable, welfare improving approximate Coarse Correlated Equilibrium computation. Finally, there is a discussion on future work which aims to generalize the developed framework. The code corresponding to the equilibria computation will be released publicly in this repository [3].

Optimizing SigmaOS for Efficient Orchestration of Fault-Tolerant, Burst-Parallel Workloads

Thu, 01 May 2025 00:00:00 GMT

Optimizing SigmaOS for Efficient Orchestration of Fault-Tolerant, Burst-Parallel Workloads Chang, Ryan SigmaOS is a multi-tenant cloud operating system designed for efficient orchestration of fault-tolerant, burst-parallel workloads. It provides users with isolated cloud environments called realms, where resources are accessed through a Unix-like filesystem interface, and supports applications built from procs—lightweight, rapidly-spawnable programs that can be both short-lived for bursty tasks or long-running and stateful for persistent services. However, the current prototype exhibits performance bottlenecks that hinder its scalability for larger, more demanding applications. This thesis addresses these limitations by introducing two key optimizations: (1) a rearchitected watch API, enhancing its efficiency and scalability for monitoring directory changes crucial for inter-proc coordination and event notification, and (2) a new ft/task server, providing a robust and high-performance mechanism for managing fault-tolerant bags of tasks, essential for applications like MapReduce. Through these enhancements, this work demonstrates significant improvements in SigmaOS’s performance on the MapReduce benchmark, showcasing improved scaling capabilities for larger cluster deployments, larger inputs, and more granular tasks. These optimizations are crucial steps towards enabling SigmaOS to effectively realize its vision as a scalable and performant platform for complex cloud workloads.

Data-Driven Modeling and Real-Time Optimal Control of Continuous Manufacturing Processes

Thu, 01 May 2025 00:00:00 GMT

Data-Driven Modeling and Real-Time Optimal Control of Continuous Manufacturing Processes Gomez, Samuel John When faced with complex disturbances, continuous manufacturing processes require robust control and adaptability to maintain product quality and operational efficiency. Although advanced control strategies such as linear quadratic regulator, model predictive control, and adaptive control have demonstrated strong performance, many industrial processes still rely predominantly on classical proportional-integral-derivative (PID) controllers because of their simplicity, ease of implementation, and sufficient results. This thesis investigates the effectiveness of data-driven modeling techniques in capturing system dynamics more accurately than traditional physics-based models. It further examines using a high-fidelity digital twin, constructed from experimental data via linear system identification and nonlinear deep learning (NARX) approaches, to optimize PID controller parameters through simulation-based gradient descent methods. A comprehensive experimental platform was developed to collect synchronized sensor and video data from a roll-to-roll continuous manufacturing system, specifically targeting disturbance scenarios that cause process interruptions. The digital twin created from these data was validated against physical experiments and shown to outperform conventional physics-based models when predicting the system’s dynamic response to disturbance inputs. Optimal control of the system was explored by implementing a virtual PID controller that closely replicates the physical controller. Optimal gain settings were identified through simulation and applied to the physical manufacturing process. The experimental results showed a significant reduction in the mean squared error and the maximum web deviation. These results demonstrate the substantial potential of digital twin-driven, data-centric control approaches in enhancing resilience, efficiency, and adaptability in manufacturing processes. This research also lays the foundation for the future development of real-time, adaptive, and autonomous control strategies in industrial applications.

Power and Progress in Japan: The Past, Present, and Future of Japan as a Tech Powerhouse

Thu, 01 May 2025 00:00:00 GMT

Power and Progress in Japan: The Past, Present, and Future of Japan as a Tech Powerhouse Maruyama, Shun This paper analyzes Japan’s economic and technological history since the Meiji Restoration through the framework of Power and Progress proposed by Acemoglu and Johnson (2023), focusing on the concepts of direction of technology and productivity bandwagon. A historical review reveals that technological progress and the distribution of its benefits were not determined solely by market mechanisms or technological inevitability, but were shaped by the power dynamics among governments, companies, workers, and others. Periods when workers held strong bargaining power and inclusive social institutions were in place saw the emergence of a virtuous cycle, in which the direction of technology moved toward broad-based innovation and the productivity bandwagon functioned effectively. Conversely, after the collapse of the bubble economy, a shift in the power balance in favor of companies led to a rise in short-term cost-cutting, resulting in a divergence from inclusiveness and innovation in the direction of technology, as well as a breakdown of the productivity bandwagon. This ultimately undermined Japan’s ability to leverage the strengths of its production system and led to a decline in technological capabilities. Currently, a new wave of technological innovation centered on AI is emerging. However, its impact remains heavily dependent on existing employment practices and corporate behavior models, making a short-term shift in direction unlikely. In the medium-to-long term, however, the societal will and collective action may create an opportunity to rebuild a virtuous cycle. This paper proposes action guidelines for companies, workers, and the government, and argues that realizing true prosperity from technological progress requires reassessing existing power structures and actively choosing new pathways as a society.

Bluespec Language Server: Adapting Rust Analyzer for Bluespec SystemVerilog

Thu, 01 May 2025 00:00:00 GMT

Bluespec Language Server: Adapting Rust Analyzer for Bluespec SystemVerilog Chan, Martin The Language Server Protocol (LSP) and popularity of VS Code have facilitated the current ubiquity of smart code editing features like hover or goto-definition. These features are powered by language servers, which are programs that perform compiler-like functions at keystroke latency on potentially incomplete code. Mainstream languages like Rust or Python have the large userbases to motivate the creation of bespoke language servers like Rust Analyzer or Pylance. However, smaller languages like Bluespec SystemVerilog, used in computer architecture classes at MIT, often need to make do without a language server. As students come to expect smart code editing features, they may miss the convenience when working with languages like Bluespec. In this thesis, we present a Bluespec Language Server forked from Rust Analyzer. This involved adapting the Rust Analyzer parser, HIR, and other internals to work for Bluespec SystemVerilog. The resulting artifact supports the full suite of typical smart editing features for classroom-grade Bluespec projects and continues to mostly work for industrial-grade projects. We discuss the many changes and challenges required to adapt this language server to work for a different language than it was designed for. Further, to address the current gap in the literature covering language server implementation, we include thorough discussion of the overall system architecture and several important subsystems with significant overlap with Rust Analyzer's internals. Finally, we conclude with a discussion of current limitations of our language server and directions for future work.

A Microelectromechanical-Cantilever Hydrogen Sensor with Palladium-Driven Bending and Piezoresistive Readout

Thu, 01 May 2025 00:00:00 GMT

A Microelectromechanical-Cantilever Hydrogen Sensor with Palladium-Driven Bending and Piezoresistive Readout Andrade, Marco A. Hydrogen gas (H₂) is considered a promising source of environmentally friendly and sustainable energy of benefit for global decarbonization. However, given the flammable and explosive nature of H₂, highly sensitive and selective detection systems with fast response are needed to enable leakage monitoring to ensure safe deployment and use. To address this need, we propose a microelectromechanical (MEMS) platform for H₂ sensing with the aim of achieving sub-1-ppm sensitivity. Our platform employs a MEMS structure that has H₂-responsive palladium (Pd) features. Once exposed to H₂, the Pd lattice expands as H₂ diffuses into it. This results in the structural deflection of a mechanically-mobile feature, in particular a cantilever. This deflection is measured using piezoresistors, which are embedded in the cantilever using a spin-on glass doping process. Piezoresistors enable rapid high-accuracy detection and quantification of H₂, as will be shown in this thesis through a combination of modeling, sensor development, sensor fabrication, and basic experimental characterization. In this thesis, we have successfully developed a fabrication plan, demonstrated the two key aspects of our fabrication, namely beam release and piezoresistor fabrication, shown beam bending driven by absorption of hydrogen by palladium, and shown that our piezoresistors respond to beam bending. Our physical results match our theoretical predictions for a beam of size 100 µm by 20 µm and a resistor with resistance 115 kΩ fabricated on SOI chips. This beam could be used to detect H₂ below 1 ppm.

Emergent Dynamics in AI-Enhanced Entrepreneurship Education: A System of Systems Study of the Orbit Tool

Thu, 01 May 2025 00:00:00 GMT

Emergent Dynamics in AI-Enhanced Entrepreneurship Education: A System of Systems Study of the Orbit Tool Dale, William The convergence of artificial intelligence and entrepreneurship education has opened a novel frontier in pedagogical innovation. The deployment of Orbit—a bespoke generative AI tool—within MIT’s 15.390 entrepreneurship course, which follows the structured Disciplined Entrepreneurship framework, is examined through a System-of-Systems perspective. This approach reveals how the tool functions not as an isolated feature but as an integrated element within a multifaceted educational ecosystem. Drawing on quantitative usage data across three consecutive academic semesters (Spring 2024-Spring 2025) complemented by course evaluation metrics, our mixed-methods approach reveals the multidimensional impact of AI-enhanced entrepreneurial education. The findings demonstrate that Orbit, particularly in its refined v2 iteration, functions as a powerful External Enabler that significantly reduces both the opacity and agency-intensity inherent in complex entrepreneurial frameworks. This enabling function manifested through measurable increases in student adoption, idea generation, and iterative engagement with critical DE steps. Beyond efficiency gains, we identify a substantive Transformation of Learning where students developed distinctly different engagement patterns—characterized by increased iteration, greater willingness to tackle complex entrepreneurial challenges, and enhanced overall course experiences. This transformation appears to deepen rather than merely accelerate learning, as evidenced by improved course evaluations alongside increased time investment in coursework. However, our analysis reveals that this transformation operates within the constraints of what we term AI’s "Jagged Frontier"—an uneven landscape of capabilities leading to differentiated impacts across DE tasks and student segments. The evolution from Orbit v1 to v2 underscores how thoughtful system design and curriculum integration critically influence the effectiveness of educational AI tools. This research contributes a nuanced understanding of how specialized AI tools can enhance entrepreneurship education while highlighting that their benefits depend on deliberate design choices, strategic pedagogical integration, and recognition of current technological limitations. The SoS framework proves instrumental in capturing these emergent dynamics, offering valuable insights for educational technologists, entrepreneurship educators, and institutions navigating the AI-enhanced learning landscape.

System Design and Evaluation of Spectrum Management Architectures for Co-Primary Sharing in the 37 GHz Band

Thu, 01 May 2025 00:00:00 GMT

System Design and Evaluation of Spectrum Management Architectures for Co-Primary Sharing in the 37 GHz Band Alsehali, Mohammed S. This thesis presents a system design framework for evaluating spectrum management architectures enabling co-primary access in the 37 GHz band. Motivated by increasing demand for mid-band and mmWave spectrum, and recent policy directions for federal-commercial sharing, this research investigates the trade-offs between utilization efficiency, coordination overhead, and interference performance across thousands of feasible spectrum management system. Using a morphological matrix, eight key architectural decisions were defined, including coordination topology, licensing mechanism, frequency planning, sensing mode, and access priority. A parametric event-driven simulation model was developed in Python to evaluate 2,808 valid architectures under low, medium, and high spectrum demand scenarios. The performance metrics, Spectrum Utilization Efficiency (SUE), Coordination Index (Cindex), and Blocking Probability (BP), were used to generate multi-dimensional tradespaces and identify Pareto-optimal solutions. Results indicate that semi-dynamic spectrum management systems with decentralized or hybrid coordination topologies consistently dominate the Pareto frontier across all demand levels. Compared to fully dynamic systems, semi-dynamic designs achieve 80–90% of the utilization efficiency with way less than 50% of the coordination cost. The results validate key hypotheses about performance trade-offs and offer actionable insights for regulators and system designers. This thesis recommends semi-dynamic, co-primary frameworks for initial 37 GHz implementation and proposes future research directions, including agent-based modeling, economic behavior integration, and acuarate physics modeling.

Architecting Innovation in Healthcare: A Case Study Review in Digital Pathology

Thu, 01 May 2025 00:00:00 GMT

Architecting Innovation in Healthcare: A Case Study Review in Digital Pathology Jezewska, Martyna The Mayo Clinic, a renowned non-profit organization, has long been at the forefront of healthcare innovation. This thesis explores the implementation of digital pathology within the Mayo Clinic, focusing on its potential to enhance diagnostic accuracy, increase efficiency, enable remote collaboration, and ultimately improve patient care. By leveraging the Architecting Innovative Enterprise Strategy (ARIES) framework, this research provides a comprehensive analysis of the socio-technical aspects of digital pathology implementation. The study begins with a literature review on innovation and its application in healthcare, followed by an in-depth case study of the Mayo Clinic's journey with digital pathology. Key findings highlight the importance of organizational design, stakeholder engagement, and continuous improvement in successfully integrating digital pathology into existing healthcare systems. The research concludes with recommendations for future innovations and insights on how healthcare institutions can better prepare for and adapt to disruptive technologies.

Optimization of Renewable Energy Siting Decisions through Vertical Axis Wind Turbine Integration

Thu, 01 May 2025 00:00:00 GMT

Optimization of Renewable Energy Siting Decisions through Vertical Axis Wind Turbine Integration Suresh, Nithyaharini The rapid increase in wind energy deployment is critical to achieving net-zero carbon emissions in the United States. However, conventional Horizontal Axis Wind Turbines (HAWTs) face deployment constraints due to their large spatial requirements stemming from their size itself and turbine spacing to accommodate wake interference. Their large footprint makes it impractical to deploy in densely populated and restricted areas, such as military zones and urban regions. This setback results in the underutilization of available wind resources, limiting wind energy’s full potential. To overcome these constraints, Vertical Axis Wind Turbines (VAWTs) offer a spatially compact alternative, enabling deployment in space-constrained areas. This study investigates the feasibility of VAWTs as a complementary wind technology by integrating them into a renewable energy siting optimization framework. This framework considers HAWTs, Solar Photovoltaics (PV), battery storage, etc., within the New England region, assuming a 100% decarbonized power system. The model utilizes an analysis that aims to minimize total system costs to assess VAWTs under varying capital expenditures and land-use restrictions. A novel feature of this study is the usage of the land availability cutoff and land restriction cases that are introduced to realistically mimic real-world land use constraints that influence wind turbine siting. The land availability cutoff defines the minimum area of land usable within the parcel for it to be considered for HAWTs and Solar PV deployment, given their larger spatial footprint. Parcels below this land cutoff are excluded from those technologies and only consider VAWTs due to the lower land available within the parcel, representing constrained regions. This methodology offers a more technical modeling of spatial constraints for renewable energy siting and allows for a realistic assessment of VAWT feasibility. Results indicate that, at current commercial costs, VAWTs are less competitive withm HAWTs and solar PV, primarily due to their early stage in the technology development and their significantly higher CAPEX, which is approximately ten times that of HAWTs. To test the technology’s viability with hypothetical utility-scale costs, where VAWT costs fall within the range of $1,300–$1,500/kW, the model still preferentially selects HAWTs due to their higher capacity factors. However, when the model considers different land use restriction cases for VAWT technology, as compared to HAWTs and Solar PVs, VAWTs become significantly more viable. VAWT placement becomes notable in these cases, increasing its share in the energy mix by 2.61% to 10.32% in favorable conditions. At high levels of land availability on a per-parcel scale, specifically, when more than 70% of the land identified as technically suitable remains available for any deployment, high-quality sites with favorable wind resources and high capacity factors continue to support HAWTs as the dominant technology given their lower Levelized Cost of Energy (LCOE). However, when the land availability cutoff increases beyond 70%, reducing siting opportunities for HAWTs and solar PV, the reliance is shifted towards VAWTs, amplifying the impact of their higher LCOE on overall system costs and making cost differentials between technologies more critical. These findings emphasize that while CAPEX reductions are critical in scaling VAWTs and driving up their competitiveness, land-use policies and spatial constraints are primary determinants of deployment feasibility. The study highlights the need for targeted policy intervention for flexible siting policies and continued research to optimize VAWT deployment strategies, ultimately enhancing wind energy integration in land constrained regions within New England and maximizing wind resource potential.

Computational Targeted Codon Optimization and Translation with Deep Learning

Thu, 01 May 2025 00:00:00 GMT

Computational Targeted Codon Optimization and Translation with Deep Learning Chemparathy, Anugrah Codon optimization—the task of recoding a protein’s underlying DNA sequence to maximize expression in a target organism—is a complicated biological optimization problem. Each gene brings a dynamic combination of local and long-range dependencies along with globally imposed constraints specific to the organism. While most existing tools for systematic codon optimization are restricted to optimizing under the constraint of a fixed amino acid sequence, recent architectural advancements in deep learning have made it possible to introduce partial modifications to the amino acid sequence without affecting protein function during the codon optimization process. Such approaches greatly increase the search space of feasible sequences, potentially opening up pathways to previously unconsidered DNA sequences with significantly greater expression rates. In this thesis, we seek to understand and improve the inverse-folding codon optimization model CodonMPNN, the behavior and performance of which have not yet been fully evaluated. We present a detailed empirical evaluation of CodonMPNN, characterizing its performance across reconstruction and translation tasks and demonstrating that it captures higher-order codon usage patterns. We produce evidence that the CodonMPNN’s training has successfully captured nontrivial aspects of the codon distribution for 1000 unique organisms, and are able to better characterize the optimal tasks that CodonMPNN’s non-synonymous nature may be able to solve. Then, by a combination of improved pretraining and a new inference-time evolutionary algorithm we are able to modestly improve the base performance of CodonMPNN from its original publication. Together, these contributions yield a measurable improvement in CodonMPNN’s practical performance and provide actionable guidance for its application in constrained codon design. More broadly, this work highlights the importance of application-aware evaluation when deploying machine learning models in synthetic biology and motivates the design of future architectures that are better aligned with real-world usage constraints.

The First Signs of Vision

Thu, 01 May 2025 00:00:00 GMT

The First Signs of Vision Chang, Cathy There has been a lot of research on the evolution of eyes through the lens of biology; however, there have been a distinct lack of research in simulating what animals saw as their eyes evolved. This project aims to create interactive simulations of the evolution of animal visions from the Cambrian Explosion to present day through the use of extended reality (XR) environments. Our goal is to communicate and educate about the evolutionary timescale to help our audience understand 1) the history of vision and intelligence and 2) how vision came to be and why it is the way it is. In addition, we want to bridge the gap between technology and vision research to help people better understand and visualize this evolutionary process. We have also collaborated with the Museum of Science and the MIT Museum to display this work in events at their venues.

The Steerability of Generative Models: Towards Bicycles for the Mind

Thu, 01 May 2025 00:00:00 GMT

The Steerability of Generative Models: Towards Bicycles for the Mind Bentley, Sarah Generative models have rapidly advanced in their ability to produce diverse, high-quality outputs. Yet their practical utility often falls short: users frequently struggle to guide models toward desired outputs, even when the model is capable of producing those outputs. This thesis argues that unlocking the full potential of generative AI requires not only improving what models can produce (producibility), but also how effectively users can guide them toward producible outputs (steerability). In short, how can we make the entire producible sets of generative models easily accessible to humans? Our contributions are fourfold. First, we formally define steerability and introduce a framework for evaluating it independently of producibility. Second, we instantiate this framework through benchmarks on the steerability of text-to-image and language models. We find that not only is steerability poor, but steering doesn’t reliably improve with more attempts. Third, we propose a framework for designing and optimizing steering mechanisms – tools that help users articulate and achieve their goals with models – and introduce Reinforcement Learning for Human Steering (RLHS) to systematically optimize these mechanisms. Finally, we instantiate this framework in a new steering mechanism for image generation that enables users to steer via images rather than text prompts. This mechanism achieves over 2x improvement over traditional text-based prompting on our benchmark. Our mathematical framework provides a generalizable path forward for measuring and improving the steerability of generative models, while our implementations of that framework empirically demonstrate its utility and viability. Overall, we define a new axis – steerability – upon which we can vastly improve generative models not only as tools for automation, but as bicycles for the mind.

Lentiviral Vector Engineering for High-Throughput Immune Profiling

Tue, 01 Feb 2022 00:00:00 GMT

Lentiviral Vector Engineering for High-Throughput Immune Profiling Dobson, Connor S. The ability to decipher immune recognition is critical to understanding a broad range of diseases, including cancer, infection, and autoimmunity, as well as for the development of countermeasures such as vaccines and immunotherapy. Efforts to do so have been hampered by a lack of technologies that are capable of scaling to simultaneously capture the complexity of the adaptive immune repertoire and the landscape of potential antigens. Each individual’s immune repertoire consists of tens of millions of unique receptors that are responsible for surveying the trillions of possible antigens that might be encountered in one’s lifetime. As a result, there has been intense focus on the development of tools for screening large antigen sets or large collections of potential immune receptors, but most of these capture complexity on only one side of the interaction. We have therefore used synthetic virology approaches to engineer a “lentivirus surface display” platform capable of screening complex antigen mixtures against the full complexity of the adaptive immune repertoire. In Chapter 2 of this thesis, we describe our molecular engineering approaches that enabled the development of VSVGmut, an efficient and modular targeted pseudotyping strategy. In Chapter 3, we leverage VSVGmut and further advances to enable one-pot library on library antigen identification screens for T cells by displaying antigens on the surface of lentiviruses and encoding their identity in the viral genome. Antigen-specific viral infection of cells allows readout of both antigen and receptor identities via single-cell sequencing. In Chapters 4 and 5, we extend our approaches to B cells and present preliminary data for applications in both cellular and humoral profiling. Taken together, our approaches represent a new class of tools for identifying the molecular targets of the adaptive immune response at scale.

Safety Analysis and Design Improvement for Semi-Automatic Train Operation (STO) in High-Speed Rail Using STPA

Thu, 01 May 2025 00:00:00 GMT

Safety Analysis and Design Improvement for Semi-Automatic Train Operation (STO) in High-Speed Rail Using STPA Suzuki, Wataru In Japan, the Tokaido Shinkansen, a major high-speed rail corridor, plans to introduce Grade of Automation 2 (GoA2) through Semi-Automatic Train Operation (STO). While partial automation promises advantages such as reduced driver’s workload and enhanced efficiency, it also creates new risks due to increasingly complex interactions among automated control systems, human operators, and physical infrastructure. This thesis aims to systematically identify and address potential hazards arising from STO in high-speed rail. By using the Tokaido Shinkansen’s announced plan as a model case, the research seeks to uncover scenarios in which normal, non-failed system behaviors can still lead to unsafe outcomes, and to propose design solutions that mitigate those risks early in development. To achieve this, the study applies Systems-Theoretic Process Analysis (STPA). Rather than isolating hardware and function failures, STPA models the entire system as a hierarchical control structure, examining each controller’s possible unsafe actions and their feedback pathways. The analysis reveals hazard scenarios that traditional failure-based methods might overlook. Examples include cases where a passenger is not detected between the train and platform doors at departure, or where verbal and signal instructions conflict and delay the driver’s response. These scenarios can happen even without any component failure. Drawing on these insights, the thesis recommends a variety of design improvements, such as new monitoring functions for subsystems, modifying instruction interfaces, and strengthening the software logic of automation systems. These findings demonstrate the value of conducting a holistic safety analysis using STPA at the conceptual design stage, before late-stage changes become more expensive. Moreover, this research provides a comprehensive, system-level railway hazard analysis, and the proposed measures can be broadly applicable to high-speed rail systems with automation.

Biomechanical Golf Swing Analysis using Markerless Three-Dimensional Skeletal Tracking through Truncation-Robust Heatmaps

Thu, 01 May 2025 00:00:00 GMT

Biomechanical Golf Swing Analysis using Markerless Three-Dimensional Skeletal Tracking through Truncation-Robust Heatmaps Taylor, Benjamin F. The efficient generation and transfer of energy in the golf swing has long been a subject of biomechanical interest, with a particular focus on the concept of the kinematic sequence, which is the coordinated segmental rotation of the pelvis, torso, arms, and club. While previous studies have modeled aspects of this sequence using high-end laboratory setups or proprietary systems, few have provided open, quantifiable, and time-resolved measurements of angular kinematics across the full swing cycle. This thesis seeks to address this gap by implementing a markerless temporal skeletal tracking approach built on the open-source MeTRAbs computer vision framework to model and measure joint angles and angular velocities throughout the golf swing. Using two-dimensional video footage of right-handed golfers performing driver swings, the MeTRAbs pose estimation model and supplemental cross-frame temporal motion sequencing code were used to reconstruct three-dimensional joint trajectories and compute rotational kinematics of key body segments. This study demonstrates the feasibility of using markerless pose estimation to extract golf swing signatures and angular velocity profiles without requiring expensive or inaccessible motion capture equipment. Preliminary analysis suggests that joint coordination patterns and temporal characteristics of body segment angular velocities may reveal quantifiable insights into the kinematic sequence, laying the groundwork for further research and instructional applications. Ultimately, this thesis contributes a replicable and cost-effective framework for analyzing golf swing biomechanics using open-source tools and computer vision.

Opportunities in Advanced Wireless Integrated Circuits

Thu, 01 May 2025 00:00:00 GMT

Opportunities in Advanced Wireless Integrated Circuits Fareed, Mo The continued evolution of wireless communications, novel compact radars, and power electronics has driven demand for high-performance semiconductor materials capable of operating at higher power density, fast switching speeds, and improved efficiency. Gallium Nitride (GaN) has emerged as a leading candidate due to its superior electrical properties compared to traditional silicon (Si), silicon carbide (SiC), and gallium arsenide (GaAs). GaN’s high power density, thermal stability, and high-frequency operation make it an ideal candidate for applications in 5G/6G infrastructure, satellite communications, defense radar, electric vehicles, and power electronics. However, widespread commercial adoption of GaN faces significant barriers, including high production costs, supply chain constraints, and integration challenges within existing silicon-based fabrication processes. This thesis explores the opportunities and challenges associated with GaN-based integrated circuits (ICs) in the context of advanced wireless systems by utilizing Dr. Eugene Fitzgerald’s innovation framework – Technology, Markets, and Implementation (TMI). A comparative analysis of monolithic vs. board-level GaN integration is conducted. The research highlights that scaling GaN wafer production to approximately 10,000 wafers per year (200mm sized wafers) is necessary to achieve cost-effective monolithic integration, yet current defense-driven demand is insufficient to drive economies of scale. Instead, commercial applications—such as telecommunications, power electronics, and consumer RF devices—are target audiences that can take advantage of monolithic integration in high volume. The findings indicate that while defense applications have led non-monolithic GaN adoption (that is, discrete GaN transistor adoption), they cannot sustain large-scale production alone due to small volume. The semiconductor industry must navigate manufacturing bottlenecks, cost reduction strategies, and foundry availability to ensure GaN’s transition from a niche, high-cost technology to a commercially viable solution. By mapping the TMI intersections and addressing economic and technical barriers, this thesis provides strategic insights into how GaN technology can achieve scalable production, unlock new market opportunities, and shape the future of advanced wireless integrated circuits.

Stock-constrained design of pseudo-standard walls from studs off-cuts

Thu, 01 May 2025 00:00:00 GMT

Stock-constrained design of pseudo-standard walls from studs off-cuts Fontaine, Anouk The AEC industry is responsible for 40% of global greenhouse gas emissions and 38% of EU waste, much of which is landfilled. The AEC waste represents an immense portion of resources that could be used instead of new materials. Many ongoing research projects have explored ways of reusing irregular components in construction, from whole steel trusses to single elements, triangulated subparts, or even irregular wood offcuts in order to mitigate the intensive recycling and deconstruction processes. However, the research has focused on general methodologies or one-off prototypes. This paper introduces a systematic approach to repurpose discarded steel and timber studs - components that make up to 10% of waste on local sites (Parigi, 2021) - into modular, steel-frame, load-bearing walls, providing a way to build new structures for the growing global demand for housing and infrastructure, while minimizing the creation of new emissions through the use of waste elements. Through a topdown and stock-constrained design approach, geometry optimization through a matching algorithm is combined with topology optimization to generate and evaluate various configurations to minimize new emissions and maximize structural efficiency. A human-scale prototype further assesses costs, architectural and structural flexibility, construction feasibility, robotic efficiency, and embodied emissions, offering a promising pathway for sustainable construction through effective waste reuse. For the available inventory, a human-scale prototype gives data on the workflow. This approach tackles the issues of existing waste stock with the growing demand for infrastructure and minimizes embodied emissions through structurally efficient resource use by pushing forward a systematic implementation of reuse in common construction practices.

Ensuring Security of Supply while Decarbonizing Islanded Heavy Industrial Electricity Systems

Thu, 01 May 2025 00:00:00 GMT

Ensuring Security of Supply while Decarbonizing Islanded Heavy Industrial Electricity Systems Kumar, Prashant Electricity is set to become the central pillar of both energy production and consumption in the global effort to achieve net-zero emissions. As key sectors—transportation, chemicals, and heavy industry—seek to decarbonize by electrifying their operations, industrialized nations face mounting strain on their electricity systems. This strain is further compounded by the rising demand for electricity driven by data centers and artificial intelligence applications, heralding a future of potentially unrelenting load growth. In such a context, it becomes not merely prudent but essential to approach decisions regard- ing investment and operation in the electricity sector with analytical rigor. Advanced capacity expansion models provide the tools for this task. In this thesis, the GenX model is employed to study Taiwan’s electricity system—an islanded, industrially-intensive grid—evaluating the evolution of its capacity mix, generation profile, prices, emissions, and overall costs. Our findings suggest that a reliable path to decarbonization lies in a considered combination of natural gas-fired generation with carbon capture and storage (CCUS), renewable sources such as solar and wind, and energy storage systems. Furthermore, this study finds that integration of nuclear and geothermal technologies significantly improves the cost-effectiveness of achieving decarbonization targets. This thesis also addresses the “missing money” problem endemic to energy-only electricity markets, examining how the introduction of a capacity market influences both investment and operational outcomes. We find that the efficacy of capacity markets is highly sensitive to the design parameters of the demand curve and the capacity credit values of the resources. For islanded systems such as Taiwan’s, a pragmatic approach to ensuring security of supply may involve retaining existing natural gas infrastructure as a strategic reserve, paired with a capacity market design that avoids excessive conservatism, leveraging the absence of policy interactions and competition with neighboring electricity markets, as observed in Europe.

Grid Enhancing Technologies: Optimization and Benefit for Distribution Grids

Thu, 01 May 2025 00:00:00 GMT

Grid Enhancing Technologies: Optimization and Benefit for Distribution Grids Anastos, Daniel One of the largest existential challenges the US and other countries face is climate change. And maybe no other system is more crucial to combatting climate change than the grid. Increasingly more requirements have been put onto the transmission and distribution grids to play an even larger role than they have in the past; consider AI, EV, residential solar, electrification of heat, decarbonization of buildings, increasing energy rates, old infrastructure. Improving the grid is a necessity to decarbonize and innovate. However, utilities, backed up by state regulation, usually, but not always, use traditional techniques to expand grid capacity and increase resiliency as opposed to investing in modern grid technology that would more quickly allow for future innovations and decarbonization. These technologies, or techniques, are broadly called grid enhancing technologies, or GETs. There are rational reasons why GETs are not used more often. Utilities are correctly, highly risk averse because they must safely and adequately supply power directly to people. Utilizing new technologies, even if proven, can be a risk that utilities are unwilling, or not allowed, to take given their role and responsibility. But these risks are largely avoided with the technologies discussed in this paper and one could argue these technologies could not only make the grid cheaper to expand but also give the grid more resilience. This paper explores how a particular grid section can increase its solar penetration by avoiding traditional hosting capacity limitations and use not even innovative GETs but GETs that are largely tested and proven. Traditionally, at some limit, the utility will stop allowing solar in an area due to various grid constraints. This paper explores how a utility may solve these constraints using new methods to avoid large grid expansion CAPEX costs and utilize new technologies or techniques. Some of the techniques explored here are commercial scale energy storage support at substations, PV curtailment, and volt-var optimization control.

Geothermal Energy Planning Considerations for Military Operational Energy Demands

Thu, 01 May 2025 00:00:00 GMT

Geothermal Energy Planning Considerations for Military Operational Energy Demands Seckfort, Cody L. Contingency locations are temporary military bases that are often established in austere or contested environments. These locations rely heavily on diesel fuel for electrical power, which creates logistical vulnerabilities and increases the risk to personnel conducting fuel resupply missions. While the Department of Defense has made progress in adopting renewable energy technologies, many of these systems remain too large, inefficient, or underdeveloped for widespread use in operational environments. Geothermal energy presents a promising but underexplored alternative for generating reliable, on-site electrical power without the need for continuous fuel resupply. This thesis evaluates the feasibility of geothermal energy systems for military operational energy demands and introduces a modified power planning process that incorporates geothermal considerations. The research focuses on closed-loop geothermal systems, utilizing an example system called the “Mil-Loop”, which is designed to minimize the system surface footprint and support remote installations. The planning process integrates existing geothermal tools, including GEOMAP/TEST for resource estimation and GEOPHIRES for system modeling and performance analysis. The Mil-Loop System Model incorporates each step of the planning process to produce a site-specific power system profile. A case study using site-specific data from Bagram Airfield was used to assess the performance of a hybrid geothermal-diesel power system. The results suggest that geothermal system integration could reduce diesel fuel consumption by up to 42.9 percent over a 40-year site lifecycle. A sensitivity analysis indicates that geothermal system power output, drilling time, and installation costs are the most critical parameters affecting system viability. Advances in drilling technology and heat extraction have the potential to reduce installation costs and timelines, making geothermal more competitive with diesel generation. This thesis also identifies a gap in military energy planning resources, specifically the lack of frameworks that include geothermal options for operational environments. It recommends that the DoD begin integrating geothermal technologies into its energy planning strategies and develop modular systems that can be deployed in contested or resource-constrained areas. While this research is limited by simplified power demand modeling and generalized tool assumptions, it offers a practical framework for evaluating geothermal viability in future defense applications. This study demonstrates that geothermal energy systems, particularly closed-loop configurations, can serve as a viable and strategically beneficial power source for military operations. When paired with targeted technology development and thoughtful integration into planning processes, geothermal systems can reduce logistical burdens, improve energy resilience, and enhance mission success in operational environments.

Levelized Cost of Fuel (LCOF) studies for microreactors using TRISO fuel in hydride and beryllium-based composite moderators in open and closed fuel cycles

Thu, 01 May 2025 00:00:00 GMT

Levelized Cost of Fuel (LCOF) studies for microreactors using TRISO fuel in hydride and beryllium-based composite moderators in open and closed fuel cycles Balla, Sai Prasad This study provides a comprehensive techno-economic evaluation of a specific class of nuclear batteries—high-temperature gas-cooled 10 MW_th microreactors (HTGRs) with TRISO fuel in prismatic- and pebble-bed cores—using four composite moderator concepts (MgO–Be, MgO–BeO, MgO–YH, MgO–ZrH). These options are compared against a prismatic graphite benchmark, under both once-through and continuous-recycle fuel cycles. In once-through prismatic systems, hydride-based moderators can reduce overall fuel-cycle costs by up to about 20% relative to graphite, whereas beryllium-based moderators may remain 40–50% costlier due to higher raw material expenses. Shifting from prismatic blocks to pebble beds decreases moderator usage and increases burnup, thus making advanced moderator options more competitive. Adopting a continuous-recycle strategy replaces enrichment with reprocessing and can further lower fuel-cycle costs by roughly 30%. Coupling a sodium-cooled fast reactor (SFR) to supply transuranic’s further reduces the cost: SFR driver fabrication and reprocessing can account for the bulk of total costs, rendering microreactor-level variations comparatively minor. Meanwhile, pebble-bed designs propose ultra-high burnups and extended residence times, which could yield significant economic gains, contingent on demonstrated long-term TRISO fuel integrity. Waste handling also factors into the analysis. Deconsolidation—removing the inert moderator before disposal—can shrink spent-fuel volumes by more than 90%, easing repository demands. Continued R&D into advanced additive manufacturing, high-burnup TRISO performance, and streamlined waste management will be crucial for capitalizing on these potential cost advantages.

Robust Dexterous Manipulation Enabled by Learning at Scale inSimulation

Thu, 01 May 2025 00:00:00 GMT

Robust Dexterous Manipulation Enabled by Learning at Scale inSimulation Bhatia, Jagdeep Singh Robots with robust bimanual dexterity have the potential to transform industries such as manufacturing and healthcare by performing complex tasks at human-level proficiency. While end-to-end learning methods have shown promise in achieving this goal, scaling these approaches remains challenging. Existing paradigms suffer from high costs associated with collecting large-scale, high-quality demonstrations on physical systems and face performance saturation due to reliance on offline data. We propose a task-agnostic pipeline that leverages robotics simulation to overcome these limitations. In particular, we introduce DART, a cost-effective, augmented reality, robot teleoperation platform for scalable data collection. We demonstrate through user study that it enables twice the throughput of existing systems. We also present a learning algorithm that integrates real-world demonstrations with reinforcement learning to surpass performance plateaus. Finally, we design a method that zero-shot transfers policies trained in simulation on real robots using only RGB input. Together, these contributions provide a practical and scalable path toward achieving general-purpose dexterous robot manipulation.

Image Registration and Gantry Tracking System of Clytia hemisphaerica

Thu, 01 May 2025 00:00:00 GMT

Image Registration and Gantry Tracking System of Clytia hemisphaerica Bunch, Bradley Understanding nervous system function and evolution requires detailed behavioral analysis of model organisms such as the jellyfish Clytia hemisphaerica. However, its size and rapid, free-swimming nature pose significant tracking challenges. This work presents a platform for the XY gantry system developed to overcome these hurdles for high-resolution behavioral monitoring. Separately, to prepare for downstream neural analysis, we developed an automated neuron segmentation pipeline - tailored for image registration purposes. Together, the tracking system and the analysis preparation pipeline provide powerful, distinct tools for high-throughput behavioral quantification and facilitate future studies linking behavior to underlying neural dynamics in Clytia hemisphaerica.

Digital Twin Technology Applied to Automotive Diagnostics

Thu, 01 May 2025 00:00:00 GMT

Digital Twin Technology Applied to Automotive Diagnostics Mwarage, Jessy Mbagara There is currently a lot of interest in the area of Digital Twin (DT) Technology. Physical product oriented organizations are increasingly looking for ways to stay ahead of the technological innovation curve in order to not get disrupted by more agile entrants. Therefore, the promise of a technology like DT is alluring for the sake of maintaining a competitive edge. This thesis seeks to explore the potential benefits of DT technology alongside what challenges might be faced in implementing one. To this end, a problem statement is formulated in the field of automotive diagnostics. This is a key value addition field for automotive companies seeking to better manage the diagnosis and repair of their automobiles in the field or the manufacturing environment. The problem is further concretized with a study of some user-driven use cases and needs in a real automotive company. From these needs, a set of requirements is formulated to guide the architecture and design of a DT demonstration. The process of architecting and designing the DT is documented. This includes a deep dive on the modeling approaches considered, the solution space for the architecture and the detailed design and implementation of a DT demonstration from a selected architectural concept. The DT demonstration is then operated under controlled conditions in order to showcase some of its capabilities. Pursuant to all this, a reflection on the effectiveness of the demonstration and the lessons learned about the implementation process are discussed. The results of the study and demonstration show some promise for organizations seeking to adopt DT technology, in this particular case for automotive diagnostics. The benefits are mainly in terms of better system architecture planning and the increased potential for better incorporating lessons learned from products operating in the field back into the design process. These benefits are weighed against the socio-technical challenges of implementing DTs from the outset of a system design exercise.

Data Acquisition for Enhancing Human-Informed Topology Optimization

Thu, 01 May 2025 00:00:00 GMT

Data Acquisition for Enhancing Human-Informed Topology Optimization Wang, Zach This thesis presents a survey application designed for the future development of HumanInformed Topology Optimization (HiTop) towards the deeper integration of optimization and real-world feasibility. Topology optimization produces high-performance designs by optimally distributing material, but its application in professional environments remains limited due to fabrication constraints and inflexible design workflows. To address this, the Carstensen Group developed HiTop, which integrates optimization algorithms with human experience, allowing engineers to modify the computer design based on their professional judgment. Thus, the future development of HiTop requires real-world data on human preferences. This project introduces a web-based survey app integrated with Qualtrics. It presents users with various design scenarios and computer-optimized designs, and records their modifications and reasoning. A preliminary survey collected responses from 13 professionals and engineering students. Preliminary findings suggest that engineers consistently focus on similar regions of interest, even when motivated by different reasons. However, the sample size is too small to make any statistically significant conclusions. While the platform mostly performed as intended, a bug related to data storage was discovered during analysis. The issue has since been resolved, and the tool is now fully functional and ready for broader deployment.

Investigating Motivational Drivers of Participation in W3C’s Web Standards Development Process

Thu, 01 May 2025 00:00:00 GMT

Investigating Motivational Drivers of Participation in W3C’s Web Standards Development Process Lauber, Emily This research investigates the motivational drivers for companies and individuals to participate in the World Wide Web Consortium’s Web standards development process. Motivational drivers are identified through a literature review, primary sources, and interviews. Thirteen semi-structured interviews were conducted with questions related to participants’ experience with the World Wide Web broadly, Web standards in general, the organization of W3C, and game modeling of the process. W3C was selected as the case study of Web-related standards bodies because of its unique model of paid membership yet open standards available royalty-free. The W3C standards process requires consensus-building, horizontal review, and proof of implementation before the organization officially recommends the specification. Existing research documents the history and value of standardization across industries, the modeling of various Standards Development Organizations (SDOs) in information industries, and the negotiation of international Internet governance. This thesis does not attempt to prove a societal benefit of Web standards but instead focuses on an individual’s belief in societal benefit and how that belief drives their engagement with W3C. Initial findings point to members seeking economic, philosophic, and moral value through participation in Web standards development. A game theory framework evaluates the economic value of different players within the ecosystem and identifies that Web browser vendors and long-time consortium members have greater power for their preferred specification outcomes than Web developers or newcomers. Despite changes in the Web ecosystem in the past 30 years, W3C members continue to be drawn to the Web for the same philosophical intents that Sir Tim Berners-Lee designed the Web for. There are shared concerns though that the economic power players identified in the game modeling has damaged or will threaten the philosophy of an open, safe, accessible Web. Interviewees shared personal beliefs that there was a moral responsibility to engage in Web standards development and enable W3C’s mission of “empowering humanity”. Further research is required to catalogue more motivational drivers, evaluate drivers across other Web-related Standards Development Organizations, and rank the priority of motivations when the different drivers are in tension.

Multi-Objective Generation of Pareto-Optimal Perception Architectures for Autonomous Robotic Systems

Thu, 01 May 2025 00:00:00 GMT

Multi-Objective Generation of Pareto-Optimal Perception Architectures for Autonomous Robotic Systems Putnam, Rachael M. Designing perception systems for autonomous robots and vehicles requires balancing sensor performance against cost, complexity, and integration constraints. This thesis introduces GO4R (Generation and Optimization of Perception System Architectures for Robotics), a multi-objective framework that jointly designs sensor selection, placement, against volumetric, entropy-based utility metric H (-) and monetary cost M ($). Perception Entropy H is formalized as a volumetric measure of uncertainty across a voxelized regions of interest (ROI), which naturally rewards coverage, overlap, and redundancy required for robust sensor fusion and calibration. NSGA-II is implemented with custom mixed-variable operators to specifically handle both continuous (e.g. sensor poses) and discrete (e.g. sensor type/count) decision variables found in this problem. Two case studies, long-range outdoor navigation on a Clearpath Jackal and short-range indoor navigation on ANYmal-C, demonstrate the framework’s ability to generate Pareto-optimal sensor architectures under vastly different ROI definitions and operating conditions. In the Jackal study, GO4R converges to a population of 11 novel Pareto-optimal designs, and revealing sensitivity to voxel size and importance weighting. In the ANYmal-C study, the compact, uniformly weighted ROI yields a flatter Pareto front with 25 Pareto-optimal designs, and underscores how intrinsic sensor parameters (e.g. angular resolution, and Field of View) dominate design trade-offs when baseline coverage is already high. Key architectural decisions are analyzed, quantified by their impact on Pareto front shape, and ordered according to the GO4R method to successively reduce uncertainty. The resulting guidelines provide practitioners with a rigorous, reusable process for tailoring perception systems to task-specific requirements. Finally, GO4R provides a publicly available NVIDIA Isaac Sim extension to aid practitioners in following the GO4R method, no matter their Autonomy application. Future work will extend GO4R to dynamic environments, improve fidelity of generated designs, and incorporate additional cost metrics such as computational load and maintainability.

Agricultural Waste Utilization: Life Cycle Assessment for Selecting Carbon-Management Best Practices on a Global Scale

Thu, 01 May 2025 00:00:00 GMT

Agricultural Waste Utilization: Life Cycle Assessment for Selecting Carbon-Management Best Practices on a Global Scale Shao, Yu-Tong Crop residues are a widely available form of agricultural waste with several possible reuse applications, including use as biofertilizers, animal feed, biofuels, and for carbon sequestration. However, in many parts of the world, large quantities of these residues are still burned in the field, releasing significant amounts of greenhouse gases (GHGs) and air pollutants to the atmosphere. This study aims to evaluate alternative and carbon-efficient strategies for reusing crop residues – especially focusing on rice straw and wheat straw – by conducting life cycle assessments (LCA) of multiple utilization pathways. Different alternative scenarios for utilizing crop residues are assessed: incorporating residue in field, animal usage for feeding, pyrolysis for electricity generation, pyrolysis for carbon sequestration, and electricity generation through residue combustion. Specifically, for the scenarios of pyrolysis and electricity generation through residue combustion, the maximum feasible transport distances of crop residues from agricultural fields to processing facilities are modeled for different logistics methods, providing information for the locations for establishing centralized facilities while maintaining GHG benefits for the scenarios. The results of this study highlight that electricity generation using crop residues, either through pyrolysis or direct residue combustion, offers the greatest climate benefits among all evaluated options. Carbon sequestration through pyrolysis also yields substantial GHG reductions, although slightly lower than the benefits from electricity generation. While crop residue-based electricity emits 4.35 to 31.25 times more GHGs per unit of electricity generated than renewable sources and 50.00 to 67.57 times more than nuclear sources, it still performs better than fossil fuels and provides added value in terms of agricultural waste management, resulting in 30.56 to 66.67% lower GHG emissions. Moreover, transportation emissions account for only a small share of the total life cycle global warming potential (GWP) in the electricity generation scenarios, ranging from 0.66% (via ships) to 16.40% (via trucks) for every 1000 km traveled. This makes long-distance residue transport viable. The findings of this study underscore the potential for crop residues to play a meaningful role in climate mitigation and sustainable agricultural waste management.

Strategic Roadmapping and Technology Portfolio Selection for Heating Decarbonization in Canada

Thu, 01 May 2025 00:00:00 GMT

Strategic Roadmapping and Technology Portfolio Selection for Heating Decarbonization in Canada Shalash, Karim Heating systems contribute significantly to Canada’s greenhouse gas emissions, accounting for approximately 117 megatons of CO₂ equivalent. demanding urgent decarbonization to meet national climate targets. This thesis employs the Advanced Technology Roadmap Architecture framework, integrating strategic roadmapping and technology portfolio selection methodologies to evaluate pathways for transitioning Canada’s heating sector to net-zero emissions by 2050. By analyzing historical emissions, forecasting adoption trends for key technologies like heat pumps, and conducting stakeholder-driven scenario analysis, this research identifies critical barriers to scaling low-carbon solutions, including high upfront costs, infrastructural limitations, and regional climatic constraints. Seven representative heating architectures—air-source heat pumps, ground-source heat pumps, district heating, hydrogen-based systems, electric resistive heating, and conventional gas-fired furnaces—are evaluated comprehensively. Among these, district heating is particularly emphasized due to its potential for significant emissions reductions and minimal consumer-bearing initial cost of ownership, especially when strategically integrated with waste heat recovery from data centers. This integration utilizes otherwise wasted thermal energy, creating a robust symbiotic opportunity for urban and industrial decarbonization. To support the practical deployment of these architectures, the thesis establishes a targeted technology portfolio comprising essential enabling and supporting technologies. Enabling technologies include centralized supervisory control systems, urban-scale district heating networks, inverter-driven compressors, advanced refrigerants, ground heat exchangers, and circulation pumps with variable frequency drives. Critical supporting technologies identified encompass building information modeling integration kits, cybersecurity modules, digital permitting platforms, smart thermostats, and thermal energy storage systems, among others. This thesis further explores technology trade-offs, focusing on structural complexity, technology readiness, and associated risks of deployment. Through detailed modeling and stakeholder-informed scenario analysis, the thesis concludes that effective decarbonization of heating in Canada necessitates substantial policy interventions, robust financial incentives, targeted infrastructure investments, and region-specific strategies. The analysis indicates that a carefully allocated $8 billion catalyst investment could close approximately 60% of Canada’s heating emissions gap by 2050. Ultimately, district heating coupled with waste heat recovery emerges as a particularly promising strategic option, underscoring its transformative potential within a diversified approach to achieving Canada’s sustainable heating future.

Proximity and Prenatal Care: Geographic Accessibility to Healthcare Facilities in N’Djamena, Chad

Thu, 01 May 2025 00:00:00 GMT

Proximity and Prenatal Care: Geographic Accessibility to Healthcare Facilities in N’Djamena, Chad Alkhalil, Kabbod Access to prenatal care is critical for reducing maternal and neonatal mortality rates. Yet, accessibility to healthcare facilities remains an understudied challenge in many sub-Saharan African countries. This study examines the spatial accessibility to healthcare facilities in N’Djamena, Chad, across various transportation modes, as well as the relationship between travel time and adherence to WHO-recommended prenatal care visits. This analysis utilized a mixed-methods approach. A geospatial analysis was conducted to estimate travel times and distances to the nearest healthcare facility across the city of N’Djamena using various transportation modes to uncover areas of low accessibility. This analysis was supplemented with survey data collected from interviews with 67 pregnant women across three different hospitals. Findings show that 72% of the surveyed population use motorcycles or cars and benefit from high accessibility. 95% of these patients have travel times under 26 and 30 minutes, respectively. In contrast, pedestrians have poor accessibility, especially when patients only attend district or national hospitals. This behavior is very likely – 81% of the surveyed population reported bypassing closer facilities, advancing familiarity and quality of care as the main reasons. In this instance, 20% of the population have travel times greater than one hour on foot. While adherence to WHO guidelines was high in early pregnancy (below 20 weeks), it declined in later stages. The study found no statistically significant correlation between travel time and adherence. Improving accessibility for pedestrians will require a combination of health system improvements, better facility distribution, and transport subsidies. The Ministry of Public Health and urban planners can employ similar data-driven approaches to plan the placement of new healthcare facilities and develop outreach strategies to ensure equitable access in a growing urban context.

Proof-of-Work Mitigation Strategy for DNS-Based Amplification Attacks

Thu, 01 May 2025 00:00:00 GMT

Proof-of-Work Mitigation Strategy for DNS-Based Amplification Attacks Bansal, Umang Distributed Denial of Service attacks, and particularly DNS Amplification attacks, have seen a steady rise in deployment over the past few decades. DNS Amplification attacks, in particular, are challenging to identify and mitigate because of their apparent similarity to legitimate DNS traffic. This thesis proposes a new Proof-of-Work mitigation strategy that provides a defense against DNS Amplification attacks and shifts the burden of mitigation to the attackers. Through our experiments, we show that our Proof-of-Work strategy is effective in reversing the impact of DNS Amplification attacks on the victim’s ability to service legitimate clients. We also provide an evaluation framework to evaluate the mitigation strategy’s impact on the victim’s quality of service.

Modeling Human-Informed Variables in Medical Data

Thu, 01 May 2025 00:00:00 GMT

Modeling Human-Informed Variables in Medical Data Abu Daoud, George In the Age of Information and Artificial Intelligence, data plays a major role in analyzing and understanding underlying trends and patterns as well as informing processes and operations. Medical data often captures information beyond mere patient conditions and state, but also human behavioral aspects of the medical process, affecting the data itself and the decisions informed by it. Modeling these variables could help us understand how they influence decisions in the field and potentially augment our models for better and more nuanced predictions. In the first study, we look into how external non-medical factors might affect decision-making by investigating the effect of 30-day mortality metrics on discharge rates following surgeries in Cardio-Vascular Intensive Care Units (CVICU) using data from the MIMIC-IV dataset. In the second study, we examine data extraction from human-notes for enhancing organ procurement decision processes.

Soil Moisture Dynamics and Thresholds for Surface Energy Balance Regime Transitions: An Observational Analysis at a U.S. Grassland Site

Thu, 01 May 2025 00:00:00 GMT

Soil Moisture Dynamics and Thresholds for Surface Energy Balance Regime Transitions: An Observational Analysis at a U.S. Grassland Site Verensia, Ria Understanding how soil moisture declines following rainfall—when the soil progressively dries due to evaporation and plant uptake—is critical for assessing plant water stress, surface energy partitioning, and land–atmosphere interactions. These periods of moisture loss, commonly referred to as soil moisture drydowns, provide a valuable window into the transition from wet to dry surface conditions. This study focuses on the critical soil moisture threshold (θ*), which marks the transition from energy- water-limited surface evaporation regimes. This transition reflects a key shift in surface energy balance and controls the extent to which evaporation is constrained by moisture availability. While previous research has typically treated θ* as a static value based on soil texture, emerging evidence suggests that it may vary depending on environmental conditions, particularly seasonal climate. This study investigates whether θ* is a fixed property or a dynamic threshold influenced by seasonal variation and available energy. Using in situ data from the Soil Temperature and Moisture Profile (STAMP) system and Infrared Thermometer (IRT) measurements at a semi-arid grassland site in Oklahoma, USA, I identify and analyze soil moisture drydown events. I estimate θ* by applying piecewise linear regression to the relationship between soil moisture and diurnal surface temperature range, isolating the breakpoint that indicates the transition from energy-limited to water-limited evaporation. Results reveal that θ* exhibits systematic temporal variations, particularly across seasons and temperature regimes, suggesting that surface temperature dynamics during drydowns are most likely a response to changes in soil moisture content. These findings challenge the assumption that θ* is solely texture-dependent and highlight the need to account for dynamic environmental controls in modeling surface energy exchange. This research provides new insights into soil moisture-temperature coupling and offers implications for land surface model development, drought forecasting, and vegetation response assessments under a changing climate.

Combustion Physics and Inverse Modeling of Energetic Materials

Wed, 01 May 2024 00:00:00 GMT

Combustion Physics and Inverse Modeling of Energetic Materials Kim, Suyong Energetic material combustion involves intricate multi-scale and multi-phase dynamics, where the interplay of chemical reaction and transport processes results in complex wave patterns across a wide range of length scales from nano to millimeter. Our limited fundamental understanding of these combustion processes poses a challenge to design and optimize combustion properties, leading to significant reliance on empirical knowledge. Deeper comprehension can be achieved by linking fundamental aspects of reaction and transport to combustion dynamics. However, there are very limited diagnostic tools available to quantify material properties and chemical kinetics for heterogeneous materials under combustion, which hinders the quantitative analysis of combustion waves. Furthermore, combustion wave dynamics and flame structures in modern nanocomposite energetic materials have not been fully resolved. This lack of breath in modeling techniques and experimental characterization has prevented quantitative analysis of combustion wave dynamics for energetic materials. This thesis aims to establish theories for combustion waves in energetic materials by correlating their intrinsic chemical reaction and transport properties with wave dynamics. To achieve this goal, two major steps are involved. First, we propose a novel inverse modeling approach to infer material properties and chemical kinetics using PDE-constrained optimization, which allows for deciphering the reaction-transport coupling from observable dynamics in currently available combustion diagnostic tools. We further discuss training challenges of neural differential equations with data subject to scale separation and propose mitigation strategies that enable learning stiff dynamical systems. Secondly, we investigate flame structures and dynamics in nanocomposite energetic materials at length scale ranging from micron to sub-millimeter using high-speed microscopic imaging techniques. Two distinct combustion wave patterns are characterized by flame dynamics and stability. Based on inverse modeling and microscopic observation, we finally construct two theories of combustion wave propagation and wave stability, by performing scaling analysis on wave dynamics in terms of mass and thermal transports and chemical reaction. A systematic view of energetic material combustion allows for deeper comprehension of how multi-scale dynamics of reaction and transports evolve in macro-scale combustion waves, potentially leading to the development of predictive models for the intricate heterogeneous combustion dynamics of energetic materials.

Computational methods for small-molecule transparent semiconductors

Sun, 01 May 2022 00:00:00 GMT

Computational methods for small-molecule transparent semiconductors Carter, Ki-Jana Solar energy has enormous potential to meet global energy demand in a renewable and environmentally sustainable manner. Although silicon-based photovoltaic (PV) devices have become significantly more affordable and accessible in recent decades, there is a need to develop alternative PV technologies which can be deployed more widely and cheaply. Visibly transparent PV devices based on organic semiconductors are well-suited to this role due to their ability to be installed on windows and building facades, their mechanical flexibility, and their high degree of tunability. However, in order for transparent PV to become commercially viable, further research is needed to shed light on the systematic tuning of the optical properties of molecular materials with visible transparency. This work applies computational tools — namely density functional theory (DFT) and graph neural networks — to gain a deeper understanding of how molecular structure impacts macroscopic optical properties and suggest directions for future study. In this work we employ linear-response time-dependent DFT with optimally tuned and screened range-separated hybrid functionals in order to compute accurate photoabsorption spectra with relatively low computational cost. Additionally, we utilize molecular graph neural networks as a means to leverage quantum mechanical datasets to accelerate the materials discovery process. These methods are combined to make progress on the optical design of organic semiconductors. This thesis document is organized as follows. Chapter 1 introduces transparent photovoltaics and the associated materials design considerations. Chapter 2 summarizes the computational methods employed in this work. Chapter 3 describes the first-principles modeling of small-molecule transparent absorbers using perylene dimide derivatives as a case study. Chapter 4 studies principles underlying the design of molecular graph neural networks. Chapter 5 applies these modeling techniques to construct a spectral dataset and train a scalable spectral model; screen a large dataset of organic molecules; identify physical trends and structure-property relationships; and suggest promising candidate materials for transparent photovoltaic applications.

MOBLLM: Model Building LLMs via Symbolic Regression and Experimental Design

Thu, 01 May 2025 00:00:00 GMT

MOBLLM: Model Building LLMs via Symbolic Regression and Experimental Design Binbas, Berkin Large language models (LLMs) have recently emerged for daily use and have already been extensively utilized for various tasks. They are shown to be able to carry out more and more complex tasks every day, including those that require a high level of formal/mathematical reasoning at human or superhuman levels. In particular, their in-context learning capabilities and the domain-specific knowledge they have via their vast pretraining corpus, as well as their fine-tunability for specific tasks drove a lot of attention and research in the field. However, applications of LLMs to the frontiers of scientific research remains an underexplored direction. In this work, we investigate how one can leverage LLMs to aid with building compact mathematical models and experimental design. Specifically, we propose a framework for using LLMs as a guide to concurrently handle the experimental design and symbolic regression tasks for data obtained from 1) a black box 1D function and 2) a black box physical system. We propose further modifications to our base framework, and perform experiments to analyze how it performs under different experiment variants, across different LLM tiers. Our experiments reveal that while larger models (of around 70b parameters) do not always achieve better downstream performance compared to smaller models (of around 8b parameters), they are able to utilize the given information and/or physical context when designing experiments and proposing symbolic expressions, and perform better than random-design baselines. We also observe that natural language constraints do not consistently improve symbolic regression accuracy. These results underscore both the challenges and the potential of integrating LLM agents into the scientific discovery process, particularly as proposers of experiments and symbolic expressions.

I-Con: A Unifying Framework for Representation Learning

Thu, 01 May 2025 00:00:00 GMT

I-Con: A Unifying Framework for Representation Learning Alshammari, Shaden As the field of representation learning grows, there has been a proliferation of different loss functions to solve different classes of problems. We introduce a single information-theoretic equation that generalizes a large collection of modern loss functions in machine learning. In particular, we introduce a framework that shows that several broad classes of machine learning methods are precisely minimizing an integrated KL divergence between two conditional distributions: the supervisory and learned representations. This viewpoint exposes a hidden information geometry underlying clustering, spectral methods, dimensionality reduction, contrastive learning, and supervised learning. This framework enables the development of new loss functions by combining successful techniques from across the literature. We not only present a wide array of proofs, connecting over 23 different approaches, but we also leverage these theoretical results to create state-of-the-art unsupervised image classifiers that achieve a +8% improvement over the prior state-of-the-art on unsupervised classification on ImageNet-1K. We also demonstrate that I-Con can be used to derive principled debiasing methods which improve contrastive representation learners.

System thinking to analyze the Market penetration of Two-Wheeled vs Four-Wheeled EVs in India

Thu, 01 May 2025 00:00:00 GMT

System thinking to analyze the Market penetration of Two-Wheeled vs Four-Wheeled EVs in India Kumbhare, Piyush This thesis analyzes the disparate market penetration rates of electric two-wheelers (E2Ws) and electric four-wheelers (E4Ws) in India, using systems thinking approaches to understand the underlying dynamics and propose strategic interventions. In 2024, while E2Ws have achieved 4.43% market penetration, E4Ws lag significantly at 1.91%, despite similar policy support. Through force field analysis and stakeholder value mapping, this research identifies key factors driving this disparity and evaluates their temporal evolution over three time horizons. The analysis reveals that E2Ws benefit from stronger driving forces, including urban suitability, favorable total cost of ownership, and simpler charging solutions, with 91% of users relying on home charging. In contrast, E4Ws face more substantial barriers, particularly in upfront costs, charging infrastructure requirements, and range anxiety. Technical modeling of key Figures of Merit (FOMs) demonstrates how different optimization challenges affect each segment's market acceptance. The research culminates in recommendations for accelerating E4W adoption, emphasizing the need for India-specific models priced similar to internal combustion engine (ICE) vehicle, localized manufacturing ecosystems, robust charging infrastructure, and innovative financing solutions. The findings suggest that while E2W adoption will continue to grow naturally, E4W penetration requires coordinated interventions across manufacturing, technology, infrastructure, policy, and consumer awareness dimensions. This research contributes to understanding how systems thinking can inform strategic planning for electric vehicle adoption in emerging markets, with specific implications for India's goal of 30% EV penetration by 2030.

Designing Generative Multi-Agent Systems for Collective Intelligence and Resilience

Thu, 01 May 2025 00:00:00 GMT

Designing Generative Multi-Agent Systems for Collective Intelligence and Resilience Dao, Nguyen Luc Large Language Models (LLMs) have been increasingly adopted by businesses to support their workflows, driving significant investment in developing generative agents. These agents can collaborate and exchange information to solve complex problems. Previous research has found that the benefits of such multi-agent systems include better performance and the potential emergence of collective intelligence characterized functionally as leadership, debate, and feedback. However, expanding multi-agent systems to include agents beyond trusted boundaries introduces the risks of malicious agents that provide incorrect or harmful information to deteriorate collective decisions or cause systemic failure. This study investigates how architectural decisions, including group size, agent prompting, and collaboration schemes, impact the system's resilience against malicious agents. Our experiment results show that increasing group size improves both accuracy and resilience at the cost of more tokens. Step-back abstraction prompting enhances accuracy and mitigates the likelihood of hallucinations induced by malicious agents. Group Chat topology is highly vulnerable to malicious interferences. Reflexion, Crowdsourcing, and Blackboard topologies offer safeguards against such risks. Eventually, we expand our research to investigate accountability gaps in generative AI systems. Designing generative multi-agent systems requires careful consideration of the trade-offs between performance, cost, resilience, and accountability.

National Space Power Analysis Through Organizational and Market Evolution

Thu, 01 May 2025 00:00:00 GMT

National Space Power Analysis Through Organizational and Market Evolution Deline, Carrie B. The space domain is undergoing fundamental changes and unprecedented growth. Once dominated by state-led missions, the space sector is now home to commercial competition, rapid innovation, and evolving models of public-private collaboration. These changes call to question how space power is built and maintained, especially during the raising geopolitical climate and power competition in space. The rise of agile commercial industry has driven down launch costs, accelerated technology development and opened new markets and business cases forcing legacy institutions to re-evaluate their strategies and business models. This thesis is motivated by the need to understand how organizations are responding to these changes, and how their choices collectively shape the United States as a national space power. Through the application of a theoretical space power model based on war strategy and Schumpeterian innovation theory, the different elements of space power will be explored in today’s context. It seeks to identify the organizational drivers of change, tensions and synergies between legacy enterprises and new entrants, and the implications of the dynamic space ecosystem. This thesis includes a mixed-methods analysis starting with a historical understanding of the evolution of the sector. By identifying current market trends, government policies and initiatives, the applied theoretical model is presented. The model is supported by market data, a force field analysis of organizational shifts, and qualitative interview insights from industry leaders. The research aims to contribute insights for government strategists and industry leaders concerned with America’s future as a space power and their organization’s role within it.

Noisy with a Chance of Mislabels: A Local and Training Dynamics Perspective on Detecting Label Noise in Deep Classification

Thu, 01 May 2025 00:00:00 GMT

Noisy with a Chance of Mislabels: A Local and Training Dynamics Perspective on Detecting Label Noise in Deep Classification Chentouf, A. Anas Noisy labels are a pervasive challenge in modern supervised learning, especially in highstakes domains such as healthcare, where model reliability is critical. Detecting and mitigating the influence of mislabeled data is essential to improving both performance and interpretability. Building on insights from training dynamics, we propose Local Consistency across Training Epochs (LoCaTE), a class of data-filtering methods that leverages over-parameterized and over-trained neural networks to distinguish clean samples from mislabeled ones. Our approach integrates both local neighborhood information and the behavior of samples across training epochs to identify noise and enhance model robustness. We evaluate our method on real (human) and synthetic label noise across three classification datasets, finding that it achieves competitive F₁ of label error detection and improved downstream accuracy using a lightweight classifier with low added computational cost.

Analyzing Multimodal Interactions through Improved Partial Information Decomposition Estimation

Thu, 01 May 2025 00:00:00 GMT

Analyzing Multimodal Interactions through Improved Partial Information Decomposition Estimation Balachandran, Adithya S. Multimodal AI aims to build comprehensive models by integrating information from diverse sensory inputs such as text, audio, and vision. However, significant challenges remain in understanding how these different modalities interact and contribute to downstream tasks. In particular, we seek to characterize how modalities complement each other, overlap in the information they convey, or contribute jointly to patterns that are not clear from any single modality alone. To address this, we propose novel methods for quantifying these multimodal interactions using information-theoretic techniques. Specifically, we will introduce a novel estimator for Partial Information Decomposition (PID) using normalizing flows, with the ability to scale well to high-dimensional data. We also develop a new framework for estimating pointwise PID, which provides insights into how individual data points contribute to information sharing and interactions across modalities, and show how to apply this framework for anomaly detection. We demonstrate the effectiveness of our methods on a variety of high-dimensional datasets, including both synthetic and real-world multimodal data such as videos.

Explanation Alignment: Quantifying the Correctness of Model Reasoning At Scale

Thu, 01 May 2025 00:00:00 GMT

Explanation Alignment: Quantifying the Correctness of Model Reasoning At Scale Bang, Hyemin To improve the reliability of machine learning models, researchers have developed metrics to measure the alignment between model saliency and human explanations. Thus far, however, these saliency-based alignment metrics have been used to conduct descriptive analyses and instance-level evaluations of models and saliency methods. To enable evaluative and comparative assessments of model alignment, we extend these metrics to compute explanation alignment — the aggregate agreement between model and human explanations. To compute explanation alignment, we aggregate saliency-based alignment metrics over many model decisions and report the result as a performance metric that quantifies how often model decisions are made for the right reasons. Through experiments on nearly 200 image classification models, multiple saliency methods, and MNIST, CelebA, and ImageNet tasks, we find that explanation alignment automatically identifies spurious correlations, such as model bias, and uncovers behavioral differences between nearly identical models. Further, we characterize the relationship between explanation alignment and model performance, evaluating the factors that impact explanation alignment and how to interpret its results in-practice.

Octavio: A Distributed System for the Sensing, Storing, and Retrieval of Piano Playing Data

Thu, 01 May 2025 00:00:00 GMT

Octavio: A Distributed System for the Sensing, Storing, and Retrieval of Piano Playing Data Abdulrezak, Ayyub MIT has a wealth of pianos spread across its campus. These instruments are owned by various groups and MIT organizations. Every day, students, faculty, and extended members of the MIT community play and practice with them. However, there currently exists no available data on their usage. This project aims to create the infrastructure for capturing this data. To this end, we installed sensing equipment on pianos across campus, constructed a matching database and filesystem of all playing sessions across time, and established a public API for the retrieval of this data. The collected data will later be used to power a publicly accessible webpage of real-time and historical visualizations, as well as serve to bolster research efforts into the characteristic piano playing of MIT.

Topology optimization of buildings-scale structures with material and fabrication constraints

Thu, 01 May 2025 00:00:00 GMT

Topology optimization of buildings-scale structures with material and fabrication constraints Jewett, Jackson L. The construction industry releases about 10% of anthropogenic Carbon Dioxide every year, primarily due to the manufacturing of construction materials. Structural optimization has been proposed as means of improving material efficiency in buildings, and thus reducing material demand for construction projects. Topology optimization has great potential for materially-efficient design because it is a free-form optimization method, allowing for performant geometries to be computationally derived with minimal input from the user. However, topology optimization algorithms must be modified to account for the specific fabrication and material constraints that are inherent in construction practices. This thesis shares a collection of research projects related to the use of topology optimization for large-scale structures relevant to the construction industry. First, a novel algorithm is proposed for large-scale 3D printed structures. The work focuses on the limitations presented by the printing nozzle, and the anisotropies that arise in 3D printed systems. Second, topology optimization is modified for design of structural glass. Several algorithms are developed, which are then used to design, fabricate, and test physical specimen to evaluate their real-world performance. Third, a framework is presented to design low-weight reinforced concrete structures. This system is used to design, build, and test reinforced concrete beams, so their performance can be compared to conventionally designed specimen. This thesis considers the diverse ways that topology optimization could be applied to design large-scale structures of various construction materials. The results demonstrate the types of computational techniques that can be used for generative design in the built environment.

Development of a 3D Printer for Oxide-free Aluminum Transportation

Thu, 01 May 2025 00:00:00 GMT

Development of a 3D Printer for Oxide-free Aluminum Transportation Smith, Henry R. Hydrogen as an energy carrier is abundant, has a high caloric value and only produces water when combusted. A challenge is directly transporting hydrogen incurs high costs due to its low density. The low flammability limit also makes it dangerous to transport. Aluminum has been proposed as an alternative energy carrier for its high density and ability to be stored at ambient conditions, allowing for cheaper transportation. Hydrogen can be produced on-site by reacting the aluminum fuel with water. However, when exposed to air aluminum forms an inert oxide layer on its surface, preventing reaction. High reaction temperatures are required to overcome the oxide layer leading to a high energy penalty. This thesis proposes a novel concept of aluminum encapsulation with water-soluble polymer. A 3d printer was designed and fabricated which creates aluminum-polymer structures that do not oxidize during storage and can achieve a wide range of reaction rates with water by varying the structure surface area. This new approach provides several benefits, by removing the oxide layer before the reaction happens, the aluminum is in an “activated” state and can react at room temperature. This reduces the energy required for reaction. Additionally, by having control over the reaction rate, ideal production rates can be achieved, reducing waste products and meeting consumption demands. The unique manufacturing flexibility of 3d printers enables the fabrication of structures with wide ranges of surface area to volume ratios. By shipping activated aluminum in the polymer structures, hydrogen can be produced locally and the need for expensive hydrogen transport can be eliminated.

Design and Development of an Atmospheric Dispersion Compensator for the LLAMAS instrument on the 6.5m Baade Magellan Telescope

Thu, 01 May 2025 00:00:00 GMT

Design and Development of an Atmospheric Dispersion Compensator for the LLAMAS instrument on the 6.5m Baade Magellan Telescope Berlanga Molina, Gerardo A. Atmospheric Dispersion is a phenomenon caused by the wavelength dependent refraction of incoming light by Earth’s atmosphere. For non-perpendicular angle of incidence, higher energy light (shorter wavelength) such as blue and violet are refracted more than its lower energy counterparts, for example, red light. As such, when a telescope is pointed at a non-zero zenith angle, there is a varying vertical angular separation between the different wavelengths of incoming light. In order to mitigate this separation and improve the spectral response of a scientific instrument, an Atmospheric Dispersion Compensator (ADC) is employed. At its simplest, this is a device consisting of two zero-deviation prisms that counter-rotate to counteract the unwanted atmospheric dispersion. At zenith, their dispersion axes cancel each other out, and near the horizon, their axes are parallel such that their net dispersion is opposite to that of the atmosphere. Here, a novel otpo-mechanical realization of an Atmospheric Dispersion Compensator is explored involving two hollow stepper motors with the appropriate diameters to house near-athermally RTV bonded powered optic lenses in order to meet dimensional constraints that prevented a more conventional ADC design. Using Hall-effect sensors, the ADC is able to reliably home without the need of motor encoders for positioning. Upon the ADC’s installation on the Large Lenslet Array MAgellan Spectrograph (LLAMAS) and the latter’s subsequent successful commissioning on the 6.5m Baade Magellan Telescope, the ADC has been operating and helping astronomers every night.

Policy and Technical Recommendations for Integrating Autonomy into Military Offensive Cyberspace Operations

Thu, 01 May 2025 00:00:00 GMT

Policy and Technical Recommendations for Integrating Autonomy into Military Offensive Cyberspace Operations Wettstein, Benjamin As AI technologies and autonomy grow, the transition to application in the military, specifically enhancing military cyberspace operations, has become both a strategic imperative and an adoption challenge. This thesis explores the challenge of effectively integrating autonomous cyber weapons systems into offensive military cyberspace operations. I offer both technical and policy recommendations to ensure autonomous technology development does not outpace its ability to be integrated. This thesis analyzes historical case studies, such as loitering munitions and escort jammers, to examine the potential for integrating autonomous cyber weapons systems into military offensive cyberspace operations. This analysis finds that the more autonomous and lethal a weapon is, the more difficult it is to integrate it into military operations. Subsequently, the current state of cyberspace operations is analyzed by discussing two cyberspace attacks, Stuxnet and Conficker. This analysis reveals that cyberspace operations currently demonstrate low to medium levels of autonomy and low levels of lethality. Therefore, there is a significant opportunity to adopt autonomous systems in the current context of offensive cyberspace operations. However, as the domain of cyberspace is transforming with the growth of complexity in technology, there are evolving legal, ethical, bureaucratic, and technical concerns. This thesis contains policy recommendations around technical standards, investment and acquisitions, and regulations regarding using autonomous cyber capabilities to address these challenges. Along with the policy recommendations, the core technical recommendation that enables autonomous cyber systems is the safe and effective deployment of human-machine interfaces to direct and control them. This thesis argues that interfaces are not merely supporting tools but are, in fact, the central technical mechanism for enabling traceability, oversight, and control in autonomous cyberspace operations. The future development and integration of autonomous cyber systems must prioritize interface design tailored to varying degrees of autonomy and operator control. The technical portion of this thesis explores different interfaces for autonomous cyber systems, utilizing distinct models of autonomy within the Cyber Operations Research Gym (CybORG) simulation environment. Each interface corresponds to the three human-machine relationships discussed, which include a semi-autonomous interface (human in the loop), a supervised autonomous interface (human on the loop), and a fully autonomous interface (human out of the loop). These interfaces serve as a proof of concept, providing evidence that different levels of autonomy can be implemented on the same autonomous cyber system. Additionally, the use of LLMs to explain the actions taken by autonomous cyber systems is explored. Ultimately, this thesis contributes technical and policy recommendations for navigating the future of autonomous cyber warfare. As autonomous systems evolve in sophistication and capability, the U.S. military must adopt policy and technical mechanisms that enable autonomy without sacrificing oversight, accountability, or effectiveness.

Response of Arabidopsis to bacterial presence under iron stress

Thu, 01 May 2025 00:00:00 GMT

Response of Arabidopsis to bacterial presence under iron stress Kitzinger, Katherine A. Iron availability is essential for the normal function of plants, but it becomes less available for uptake under drought. A lack of iron can lead to early senescence, fewer and less nutritious crops, and in extreme cases, plant death. In response to these stressful conditions, microbial interactions can lead to improved plant health, however the mechanism by which this occurs is not understood. In this study we cocultured an Arabidopsis MTP8 knockout line, which is susceptible to iron stress, as well as a subset of a previously established synthetic microbial community derived from healthy Arabidopsis roots. We cocultured the Arabidopsis lines and bacteria under three different iron levels in a hydroponics system and measured the dry weight and chlorophyll content ten days post inoculation. This study aims to narrow the potential mechanism of the beneficial effects of bacteria on plants experiencing nutrient stress.

Interactive Topology Optimization with Hybrid Truss and Continuum Elements Types

Thu, 01 May 2025 00:00:00 GMT

Interactive Topology Optimization with Hybrid Truss and Continuum Elements Types Zhang, Eileen Topology optimization is a rising tool in structural design that can improve material efficiency and promote sustainability. However, currently topology optimization is not greatly used in the industry due to the nature of its user-unfriendly, high computation cost and difficulty in manufacturability. This thesis proposes a new framework combining traditional discrete topology optimization with truss elements and continuum elements topology optimization in creating a more informed algorithm suitable for more practical design scenarios. In addition, the drawing toolkit is also introduced in helping users better interact with the system in outputting their desired outcome. The hybrid element type topology optimization is achieved by creating separate local stiffness matrices and mapping them respectively to the same global design space to perform optimization together. The interactive drawing functions are used as add-in truss members that users can select the amount and draw in the length and locations of them in the design space. This framework is tested on multiple topology optimization classic problems including cantilever beam with bracings and MBB beam. All draw-in truss hybrid topology optimized results show a more efficient design results with lower compliance and overall lower material quantity.

G-Code Based Toolpath Simulation for Predicting CNC Energy Consumption

Thu, 01 May 2025 00:00:00 GMT

G-Code Based Toolpath Simulation for Predicting CNC Energy Consumption Anziani, Jonathan Machining is an energy intensive process, and being able to model the energy consumption of machining would allow manufacturers to consider how to reduce their energy footprint. While many models have been developed for estimating energy consumption, they are not easily applicable or accessible to CNC machining, where the material removal rate is variable. This thesis develops a G-code based simulation that uses a voxel mesh to virtually recreate material removal, approximating the material removal rate at discretized points in the machining process. Using an energy consumption model and machine power data, material removal rates are related to the power consumption of machining the part. The simulation pipeline was validated using power data collected from literature, and for a constant material removal rate the model has shown average absolute error of 3.17% predicting power and 2.89% predicting specific energy consumption for simulated test geometries.

Analyzing Inconsistent Results of Table Transformer for Improved Data Extraction in Childhood Obesity Intervention Literature

Thu, 01 May 2025 00:00:00 GMT

Analyzing Inconsistent Results of Table Transformer for Improved Data Extraction in Childhood Obesity Intervention Literature Neupane, Pragya Tables in scientific literature are rich sources of structured data, yet their complex and variable formats pose challenges for automated extraction. This thesis focuses on improving the reliability of Table Structure Recognition (TSR) using the Table Transformer (TATR) model, with a specific application to childhood obesity intervention studies. While fine-tuning TATR on a domain-specific dataset improves detection metrics, persistent errors such as overlapping rows and misclassified header columns remain. Through a systematic post-hoc error analysis of 175 scientific tables, we identify these dominant failure modes and develop lightweight post-processing modules: an overlap-aware row filtering algorithm and an OCR-enhanced column boundary correction method. Importantly, instead of relying on computationally expensive large language models (LLMs), this approach leverages efficient, interpretable techniques tailored to the domain-specific structure of public health tables. Our combined method reduces the proportion of structurally erroneous tables from 46.3% to an estimated 9.7–12.6%, improving the semantic alignment and interpretability of model outputs. This work contributes a transparent and scalable pipeline that enhances the trustworthiness of automated table extraction systems, with direct relevance to evidence-based decision-making in public health.

Assessing Opportunities to Reduce Carbon Dioxide Emissions from Electric Arc Furnace Steelmaking in the United States

Thu, 01 May 2025 00:00:00 GMT

Assessing Opportunities to Reduce Carbon Dioxide Emissions from Electric Arc Furnace Steelmaking in the United States Colcord, Christopher C. Steel is energy- and CO₂ emissions-intensive to produce, but it is also a crucial material for infrastructure, defense, and the energy transition. This thesis focuses on Electric Arc Furnace (EAF) steelmaking, which accounts for roughly 70% of steel production in the United States. Decarbonization levers for EAF producers are diverse—encompassing energy efficiency (EE) measures, fuel switching, material input substitution, development of onsite carbon-free electricity (CFE) generation, CFE procurement through power purchase agreements (PPAs) or unbundled renewable energy credits (RECs), and negative-emissions credit purchases, among others. We first construct a techno-economic model that analyzes costs and emissions of individual EAF facilities in the United States under a business-as-usual (BAU) scenario for the years 2025 through 2035. We then calculate the Levelized Cost of Carbon Abatement (LCCA) of various decarbonization levers against the BAU counterfactual. We build aggregate LCCA curves to draw insights on least-cost emissions abatement strategies for facilities and opportunities for policy to accelerate decarbonization decisions. We find that the modeled levers collectively deliver a 46% reduction in EAF CO₂ emissions versus the BAU case—equivalent to a reduction of roughly 1.7% of national industrial CO₂ emissions. Voluntary CFE procurement has the greatest potential to abate EAF emissions, but comes with large uncertainties. Onsite CFE and PPAs have negative LCCAs in most cases, whereas unbundled RECs have positive LCCAs. EE measures provide modest emissions reductions and costs are negative on a levelized basis under a wide range of assumptions. EE opportunities, onsite CFE, and PPAs may be bound by non-financial constraints. Direct reduced iron (DRI) with carbon capture has lower variable costs and produces fewer emissions versus hydrogen-based DRI in most cases. While the challenges to decarbonize EAF steelmaking are immense, we find EAF facilities can take actionable steps in the near term—supported by federal and state policies—to abate carbon emissions while reducing levelized costs.

Investigating the Capacity of Generative AI to Learn Genotype-by-Environment Interactions in Brachypodium distachyon

Thu, 01 May 2025 00:00:00 GMT

Investigating the Capacity of Generative AI to Learn Genotype-by-Environment Interactions in Brachypodium distachyon Neufeldt, Charlie Climate change exacerbates environmental stressors such as drought, challenging the resilience of agricultural systems and highlighting the need to understand plant genomic architecture and its responses to such environmental variation. A key molecular mechanism underlying these responses is transcriptional plasticity: environment-induced changes in gene expression that vary among genotypes, representing one way that genotype-by-environment (GxE) interactions manifest at the molecular level. While transcriptomic data offers a unique and powerful view into these responses, traditional modeling approaches often rely on linear assumptions, limiting their ability to detect complex, nonlinear patterns of regulation. This thesis investigates whether generative machine learning modeling, specifically the use of transformers, can extract biologically meaningful representations of gene expression dynamics in plants. Inspired by the successes of the scGPT model for human genomics, I developed and trained a compact transformer architecture, the PlantGeneEncoder, on bulk RNA-seq data from two natural accessions of Brachypodium distachyon grown under drought and control conditions. The model was trained on binned expression values using both a baseline configuration and a set of regularized variants incorporating noise injection, co-expression preservation, entropy-based sample weighting, and masked gene modeling as a self-supervised objective. While baseline models achieved perfect reconstruction accuracy, they failed to preserve meaningful biological structure in the latent space. Regularized models achieved a better trade-off, maintaining high reconstruction fidelity while demonstrating improved genotype classification performance and modestly better alignment with the original expression structure. However, environmental condition signals remained difficult to capture across all configurations, with classification accuracies only marginally above random chance. These findings highlight the promise and limitations of transformer-based generative modeling for plant transcriptomics and provide a flexible framework for future efforts to model transcriptional plasticity and regulatory responses to environmental stress.

Active Multi-View Object Pose Estimation Using a Mobile Ground Robot

Thu, 01 May 2025 00:00:00 GMT

Active Multi-View Object Pose Estimation Using a Mobile Ground Robot Wynia, Ethan Accurate 6-DoF object pose estimation remains a central challenge in robotic perception, particularly when relying on single-view observations subject to occlusions and limited geometric cues. This thesis presents a system that incrementally refines object pose estimates by collecting multi-view observations with a mobile ground robot, the Clearpath Jackal. The robot autonomously navigates in a circular trajectory around a target object, capturing images while maintaining a fixed orientation toward the object center. At each waypoint, 2D image corners are manually annotated and paired with corresponding 3D object coordinates. The Perspective-n-Point (solvePnP) algorithm is then applied to estimate the object's pose relative to the camera. The system transforms these camera-centric poses into a consistent global frame using Robot Operating Systems’ transform library. Using these poses, the system tracks reprojection error to evaluate pose confidence. Across multiple trials, the mean reprojection error consistently decreased as more views were added, confirming that spatially diverse observations improve pose estimation accuracy. A cross-run analysis shows reproducible trends, with error reductions of over 40% in many cases. These results validate the efficacy of active multi-view collection for reducing uncertainty and lay the foundation for future extensions with automated key point detection and You Only Look Once (YOLO) supported multi object detection.

Design of an FSAE Cooling System for Electric In-hub Motors

Thu, 01 May 2025 00:00:00 GMT

Design of an FSAE Cooling System for Electric In-hub Motors Lohawala, Sehar This thesis explores the design of water cooled electric in-hub motors for use in high power automotive applications such as Formula SAE Electric student racecars. The main components of the drivetrain of the 4WD electric vehicle are the motor and the motor controller. This thesis focuses on designing a reliable cooling system for the motors to ensure that they operate in an optimal temperature range that increases drivetrain efficiency, prevents catastrophic motor damage, and ultimately improves vehicle performance. During the design process, extensive heat transfer analysis was conducted for liquid cooling, air cooling, and heat pipes. Additionally, CFD was conducted for various water cooling architectures to determine the influence of coolant flow direction and channel dimensions on cooling performance. Results were subsequently analyzed, plotted, and used in the motor cooling architecture selection process. After determining the motor cooling sleeve architecture and dimensions, detailed CAD was created for both plastic prototype sleeves as well as metal 3D printed sleeves. Testing demonstrated that the cooling sleeve successfully met its target pressure drop of 3-4PSI. More thorough testing and data collection of the cooling sleeve’s thermal performance is still in progress. While details on the CAD, prototyping, testing & validation, and manufacturing, are outside the scope of this thesis, images are included for reference.

Open-IRIS: Low Cost, Fully Open Source In-situ Infrared Inspection of Silicon

Thu, 01 May 2025 00:00:00 GMT

Open-IRIS: Low Cost, Fully Open Source In-situ Infrared Inspection of Silicon Becker, Aaron M. Short-Wave Infrared (SWIR) imaging has become a powerful and well-known technique over the last two decades for silicon inspection and imaging. Open-IRIS is a low-cost, fully open-source system for in-situ InfraRed Inspection of Silicon devices (IRIS). It is designed to lower the cost barrier for academic and research users requiring high-precision IR imaging of silicon microelectronics. This thesis details the design and implementation of the Open-IRIS platform, including its optomechanical components, motion control system, electrical system and software architecture. Its design is highly modular and low cost, making it an invaluable and extensible tool for many future applications, including microarchitectural security research, chip failure analysis, and biological imaging. Key design challenges, such as achieving high mechanical and optical resolution on a budget are addressed. Computational microscopy techniques, including Fourier Ptychographic Microscopy (FPM), are evaluated to improve resolution. The system’s imaging resolution on a standard resolution target is evaluated, as well as its motion repeatability and accuracy. Results show that Open-IRIS achieves 5.34 μm optical resolution with a 5x objective, and 3.47 μm resolution with a 20x objective. Mechanically, it has 6.5 μm repeatability, and 35.5 μm accuracy, all on a total budget of less then US$1000 - a fraction of the cost of comparable commercial systems. The complete design is fully open-source, enabling broader access to advanced chip inspection techniques, and serves as an excellent starting point for future expansion into advanced security research like laser fault injection.

Electrifying a Jet Ski: Designing & Manufacturing an Electric Drivetrain for a Personal Watercraft

Thu, 01 May 2025 00:00:00 GMT

Electrifying a Jet Ski: Designing & Manufacturing an Electric Drivetrain for a Personal Watercraft Hudspeth, Blake H. While many vehicles are well into their transition to electrification, marine vessels are lagging behind; specifically, there are very few electric personal watercrafts (PWC) on the market. Though many engineers and hobbyists have retrofitted combustion engines for cars, motorcycles, and other automobiles for electric propulsion, there are almost no examples of electric conversions for watercraft. This thesis details the designing and manufacturing process of an electric drive train into a 1997 Yamaha Wave Venture 760 Personal Watercraft (PWC). This project aims to serve as a proof of concept for interchanging a combustion engine from an older PWC with an electric motor and Lithium-Ion battery. I performed rigorous calculations to properly size a battery and motor to propel the watercraft. Upon learning to weld, build battery packs, and configure a battery management system, I prototyped extensively—iterating upon battery packs and pack configurations. I assembled a final 72V, 10.4 AH battery pack with waterproof housing to power the 5kW motor that I coupled to the existing impeller drive shaft. Lastly, I performed a dry test monitoring the battery and motor health, confirming successful retrofit of the electric drive train, proving the feasibility in electrifying a used personal watercraft.

Numerical Analysis of Human-Informed Topology Optimized Lateral-Load-Resisting Systems of Tall Buildings under Seismic Excitation

Thu, 01 May 2025 00:00:00 GMT

Numerical Analysis of Human-Informed Topology Optimized Lateral-Load-Resisting Systems of Tall Buildings under Seismic Excitation Blaze, Edie In the construction industry, structural, architectural, and environmental considerations can often be at odds with each other, leading to inefficient structures and, consequently, material waste. Topology optimization has shown promise as one potential solution to this problem, offering designs that are both structurally efficient and aesthetically interesting. However, topology-optimized designs are often difficult to manufacture or do not take into consideration other aspects that are crucial in the construction industry. Human-informed topology optimization, or HiTop, is a previously-developed algorithm that allows users to edit areas of interest, providing a computationally-efficient solution to address concerns with the designs. This paper uses MATLAB to apply HiTop to the design of the lateral-load-resisting systems of tall buildings, comparing results to those of three other designs: a “human” design with standard cross bracing, a optimized design using classical topology optimization, and a previously-developed algorithm which optimizes designs under a sum of modal compliances formulation, similar to how structures are analyzed in seismic codes. The designs are evaluated quantitatively, comparing natural periods, modal displacements, sum of modal compliances using modal decomposition, as well as computation time. They are also evaluated qualitatively, as HiTop is used to modify designs to improve constructability and aesthetics. The HiTop algorithm successfully created manufacturable, aesthetic designs in line with the user’s goals across a range of H/B ratios within a brief time frame. HiTop designs also performed similarly to the classically optimized designs, indicating that modifications to an optimized design to improve manufacturability, aesthetics, or other potential goals of a user do not significantly decrease structural performance under seismic loading.

Peatland burning identification among other wildfires across different ecozones in Canada

Thu, 01 May 2025 00:00:00 GMT

Peatland burning identification among other wildfires across different ecozones in Canada Chen, Ming The unprecedented severity of the 2023 Canadian wildfires highlights growing concerns about the vulnerability of global peatlands—key ecosystems storing substantial amounts of terrestrial carbon. Peatlands, traditionally resistant to burning, are increasingly at risk due to climate-induced warmer and drier conditions. This study specifically investigates the extent and characteristics of peat burning in the 2023 Canadian wildfires based on available remote sensing data. The primary objective is to determine whether fires on peatlands demonstrate distinct fire behavior compared to fires on non-peatland. To achieve this goal, this study utilized statistical tools and machine learning algorithms, including power-law relationship estimates, Mann-Whitney U test, K-means clustering, and generalized additive model (GAM) to identify the contribution of peat presence to fire behaviors. Key findings demonstrate that fires on peatland are significantly more intense, longer-lasting, and associated with higher carbon emissions. Even though peat combustion can not be confirmed without field validations, these results underscore the critical importance of the potential impact of peat on wildfire growth and management. By highlighting the disproportionate impact of peat burning, this study provides a foundation for future research aimed at developing targeted remote sensing techniques and policy responses to mitigate peatland vulnerability and preserve vital carbon stores in the context of global climate change.

The effects of different stiffnesses and coefficient of drag on the performance of a Padel racket

Thu, 01 May 2025 00:00:00 GMT

The effects of different stiffnesses and coefficient of drag on the performance of a Padel racket Mora Armendariz, Francisco David The sport of padel is the fastest growing sport in the world. As the sport has evolved there have been more and more innovations in the equipment of Padel. One of the most important was moving from wooden paddles to ones made of fiber glass and carbon fiber. As such a prototype racket was designed and manufactured for this study through experiments to obtain data on existing commercially available rackets, a specific design was proposed and created. The prototype created is then compared to the other rackets in the same way that the original properties were measured. Although this racket needs to continue to be iterated on, the potential of a viable competitive racket is promising.

Tension Activated Kirigami Structures for Interlayer Use in Circular Construction Applications

Thu, 01 May 2025 00:00:00 GMT

Tension Activated Kirigami Structures for Interlayer Use in Circular Construction Applications Bigler, Thomas Motivated by the use of fully circular materials and improving sustainability, this thesis investigates the use of tension-activated kirigami as an architected material, for which mechanical properties are determined by their designed structure (geometry) rather than the bulk material itself. These architected materials can be used as a dry replacement for adhesives in construction with glass masonry units, enabling reclaimability and recyclability of both the interlayer material and glass masonry unit. The kirigami design and material selection allow for the customization of architected-material properties for best compatibility with the glass units. The research has involved both analytical/mathematical modeling for early material and design selection and an experimental process to develop an empirical database. Experimentation on different materials, designs, aspect ratios, etc. has provided data to begin extrapolating trends and behaviors of the architected material. These data will allow for design decisions and material selections based on the functional requirements of a specific structure or application.

Sustainable Engineering of Polyethylene Fiber Materials: Advancing Functional Properties of Diverse Textile-Based Structures

Thu, 01 May 2025 00:00:00 GMT

Sustainable Engineering of Polyethylene Fiber Materials: Advancing Functional Properties of Diverse Textile-Based Structures Huynh, Amy This thesis explores pathways to circularity for polyethylene-based textiles through an integrated framework that combines material experimentation, systems-level policy analysis, and cultural innovation. Focusing on olefin block copolymer (OBC) filaments—engineered with semicrystalline polyethylene hard segments and elastomeric soft blocks—the study evaluates their mechanical behavior across a range of stitch-based textile geometries. Cyclic and postfatigue tensile testing reveals how formulation and structure shape energy dissipation and durability, informing design strategies for high-performance applications such as intra-vehicular spacesuits and wearable technologies. To understand the broader systems context, the thesis analyzes barriers to integrating recycled polyethylene (rPE) into textile supply chains, identifying economic, legal, institutional, technological, firm-level, and societal constraints. It proposes targeted strategies based on global policy trends, EU case studies, and a geospatial analysis of U.S. recycling infrastructure. Finally, the work explores how generative AI can revitalize traditional craft practices—such as bobbin lace—by co-creating patterns designed for both aesthetic and functional performance in new materials. Together, these efforts propose a model for advancing sustainable textile innovation that bridges material science, circular design, and policy transformation.

Experimental Design in Operations

Thu, 01 May 2025 00:00:00 GMT

Experimental Design in Operations Wang, Chonghuan Experimental design has been fundamental to many fields, yet applications in operations research (OR) and operations management (OM) bring in complexities, as well as opportunities, that extend beyond classical statistical goals. This thesis discusses why OR/OM should care about experimentation and its design, where the challenges lie in operational and service systems for classical experimental design, and why OR/OM researchers are uniquely suited to address the challenges. More specifically, this thesis advances experimental design by introducing more operational perspectives, addressing two core challenges: incorporating operational objectives and leveraging operational modeling to enhance experimentation. First, traditional experimental approaches, such as A/B testing, primarily aim at statistical efficiency (e.g., reducing variance or bias). However, OR/OM applications frequently involve additional operational considerations, such as welfare preservation, revenue optimization, risk control, and non-stationarity. We investigate these settings in Chapters 2–4, developing frameworks for multi-objective experimental design. In Chapter 2, we introduce a minimax multi-objective optimization formulation to balance statistical power and welfare loss, derive necessary and sufficient conditions for Pareto optimal solutions, and propose robust multi-armed bandit designs. Chapter 3 extends this approach to pricing experiments, exploring trade-offs between estimating causal effects (price elasticity), maximizing revenue, and controlling tail risks, along with robust statistical inference methods. Chapter 4 addresses non-stationary experimental environments where treatment effects dynamically evolve, designing experiments that optimally balance accurate estimation of changing effects and welfare minimization. Furthermore, we highlight the substantial value of operational models—particularly Markov Decision Processes (MDPs)—in experimental design. In Chapter 5, we address the challenge of estimating long-term cumulative outcomes, such as customer lifetime value, using short-term experimental data. We develop novel inference methods grounded in MDPs, which effectively bridge short-term data to long-term outcomes. Moreover, by recognizing many real-world treatments tend to be localized for practical efficiency, we introduce novel estimators that leverage the localized structures to achieve substantial variance reductions. In summary, this thesis underscores how OR/OM contexts uniquely enrich experimental design, offering robust theoretical frameworks and practical solutions to operational challenges, ultimately broadening both the theoretical foundations and the practical impacts of experimentation.

Fueling Conflict: A Global Dataset of Energy Protests

Thu, 01 May 2025 00:00:00 GMT

Fueling Conflict: A Global Dataset of Energy Protests Harrison, Ethan How do popular grievances about (the lack of) access to energy lead to political violence and instability? I use a mixed-methods approach to answer this question, based on a qualitative case study in Sri Lanka and a quantitative framework for tracking energy protests worldwide. Specifically, through an analysis of the 2022 Aragalaya protest movement in Sri Lanka, I elaborate on how breakdowns in state capacity to provide energy to its citizens can trigger civilian unrest. Building on this case study, as well as insights from the empirical literature on the drivers of instability related to energy access, I then pilot a machine learning (ML) framework to identify energy-related protest events in the Armed Conflict Events Database (ACLED) based on context-specific keywords, which results in the creation of the first global dataset on energy protests. This novel source of evidence, in turn, will open new avenues for research on the conflict-energy nexus, particularly on the impact of market shocks on civilian unrest and instability in low- and middle-income countries – a topic for which current empirical work is limited. I show how the ML framework I develop here can be used to enable continuous monitoring of protest activity related to energy access, as well as how the framework can be extended to other forms of political violence, offering a promising tool for peace-building initiatives across contexts. Therefore, such a framework could inform key evidence to support policymakers, practitioners, and researchers in the design of strategic policies that facilitate the provision of energy while mitigating the risk of conflict and instability worldwide, particularly in "energy-poor" countries.

Towards a Circular Lunar Economy: Incentivizing the Design of Multi-Purpose Reusable Lunar Landers and Rovers

Thu, 01 May 2025 00:00:00 GMT

Towards a Circular Lunar Economy: Incentivizing the Design of Multi-Purpose Reusable Lunar Landers and Rovers Khan, Nadia Rehman Both NASA and ESA have committed to establishing a lasting presence on the Moon by 2030. However, lunar surface debris has already exceeded 200,000 kg, prompting concerns about the environmental, operational, and economic viability of future missions. This thesis proposes that circular economy principles—particularly reusability, modularity, and interoperability—must be embedded in early mission architecture to reduce waste, improve system longevity. To evaluate these goals, this thesis introduced a novel decision-support framework, the Lunar Exploration Impact Assessment (LEIA), alongside a policy-informed set of Lunar Surface Sustainability Guidelines (LSSG). Both decision support tools were designed help mission designers and space policy stakeholders to incentivize the design of resilient reusable lunar landers and rovers. In this thesis the LEIA framework, was applied to two case studies: NASA JPL’s EnduranceA autonomous lunar sample return rover, and ESA’s multi purpose Argonaut lander to evaluate the sustainability of each spacecraft after the EOL/M phase of each mission. Scores were computed using a Multi Criteria Decision Analysis (MCDA) approach. Seven Impact Assessment Indicators (IAI)s were considered, to assign a sustainability rating for each mission: Cost-effectiveness, environmental impact, science value, redundancy, resilience, strategic value, and technological feasibility. The Endurance-A mission achieved a sustainability score of 66.4%, based on a sample collection post primary mission scenario, indicating moderate sustainability across some categories such as cost-effectiveness 18.9% and technological feasibility 12%. However, the environmental impact score was limited to 7.7%, due to the out-gassing and launch emissions associated with the SpaceX Starship lander. The rovers redundancy and maintainability ratings also constrained the overall sustainability rating – highlighting a gap in the availability of tools suitable for EVA-based repairs on the lunar surface. Subsystems most at risk of degradation—mobility, thermal, and power—require enhanced design for long-term reuse scenarios. Each of these factors were made salient through the Argonaut case study, indicating that in the short to medium term in order to prevent the accumulation of lunar surface debris lunar rovers and landers must be designed to be more resilient to the conditions of the lunar environment. To supplement the LEIA framework, a set of policy recommendations were developed in order to address the lack of End of Life (EOL) procedures in place to manage lunar surface debris – in the form of retired lunar missions. The guidelines detailed how economic policy mechanisms adopted in circular economy systems could be leveraged to incentivize the design of sustainable lunar surface missions and operations.

Beam Mechanism Failure in Multistory Steel Frame Structures

Thu, 01 May 2025 00:00:00 GMT

Beam Mechanism Failure in Multistory Steel Frame Structures Hashbarger, Brad Engineers must ensure that building structures are not in danger of collapse, so analyses always include safety factors that create redundant yet materially inefficient buildings. This has been common practice for most of structural history, but today, growing concerns for carbon emissions force designers to cut material usage while retaining the same level of safety. Processes opt into one of two processes: an overall lighter unit or stiffening specific internal systems to encourage a load path. The problem with either of these options lies in progressive collapse in the event of structural damage. If one column is lost, stresses propagate either until equilibrium or a larger collapse occurs. Progressive collapse remains a popular research area to identify specific vulnerabilities, often with numerical models for a visualization of each stress state and redundant capacity. Previous studies used analytical and experimental performance to observe the critical effects of losing an external versus internal column and the role of other components, such as joints, joists, and composite slabs, to carry additional loads. However, designs and analyses are bound by assumptions that govern model behavior. To understand the sensitivity and limits of these assumptions, this thesis predicts the performance of steel moment-frame structures of varying bay geometries, proposing deflection fields to inform modern practice in all phases of project development. Instead of numerical simulations, the process follows an analytical approach based in the fundamental methods of equilibrium and the conservation of work and energy. By designing sections for their elastic capacity, their operational performance is directly linked to their failure response. This suggests the dominance of design preferences in stability, even with changes in beam spans or floor loading. Results support an optimal span ratio for plasticity under two-way load distributions that favors bay geometry ratios (L1/L2) between 1 and 2 but varies based on failure locations and how many columns have been lost. This also emphasizes the weaknesses out of plane as span ratios range from 0.5 to 1. Project layouts can utilize the free strength provided by bay geometries as part of the structural design process. If large deflections or span lengths are expected, beam depth and section thickness should increase together to ensure beams utilize their full plastic capacity to achieve additional redundancy from catenary action. Overall, the thesis demonstrates that such considerations in the early design stage can enable steel structures to achieve greater safety with less material.

The Design and Fabrication of a Punch and Die System for 2.008 Thermoformed Parts

Thu, 01 May 2025 00:00:00 GMT

The Design and Fabrication of a Punch and Die System for 2.008 Thermoformed Parts De Jesus, Sebastian As a student in 2.008, Manufacturing and Design II, our team successfully manufactured 100 identical yo-yos. Although the class is very well structured and the CNC milling, injection molding, and thermoforming in the Laboratory for Manufacturing and Productivity (LMP) were all optimized for the class, the punch and die system was one process that was more tedious than the rest. Punches, a die, and a calibration piece were designed, fabricated, and tested to find the best clearance size and fill in the gap in documentation of punching plastics. A new, working system was successfully fabricated and assembled, and clearance size 5% was determined to have a lower margin of alignment error. The new punch system will be implemented in the LMP and used by 2.008 students.

Estimating Aboveground Biomass (AGB) Throughout the Pacific

Thu, 01 May 2025 00:00:00 GMT

Estimating Aboveground Biomass (AGB) Throughout the Pacific Domingo-Kameʻenui, Joy P. Aboveground biomass (AGB) is a signiﬁcant carbon pool in forests, making AGB a good indicator of forest health and carbon storage. AGB has been studied on multiple scales, in which allometric equations were developed to ﬁnd relationships between AGB and tree parameters. However, despite the presence of AGB studies for speciﬁc sites in the Paciﬁc Islands, there is a lack of AGB comparative studies or data syntheses focused on the Paciﬁc Islands as a whole. This study synthesized data on AGB, tree height H, land cover, and Paciﬁc Island forest community to develop allometric equations using linear and polynomial regression models for trees in the Paciﬁc based on H as the main parameter. This study found polynomial relationships between AGB and H for shrub and herbaceous covers. Speciﬁcally, AGB = 1.76 H^2 + -51.01 H + 346.53 for shrub cover (adjusted R^2 = 0.94, n = 39), and AGB = 1.11 H^2 + -81.97 H + 1167.20 for herbaceous cover (adjusted R^2 = 0.71, n = 79). However, future research and data collection would be necessary to develop allometric equations for tree cover and barren land cover. No signiﬁcant correlation was found between AGB and H for Paciﬁc Island forest community. This study may help with forest management and conservation practices, along with carbon sequestration and storage practices in forests, in the Paciﬁc Islands. This study may also contribute to Paciﬁc-led climate change mitigation and adaptation methods and initiatives.

AI-powered Data Mining for the Development of Sustainable Concrete Materials

Thu, 01 May 2025 00:00:00 GMT

AI-powered Data Mining for the Development of Sustainable Concrete Materials Duan, Yifei Data mining has become essential to contemporary industrial and scientific research, playing a pivotal role in uncovering insights from large-scale industrial datasets and literature collections. The sustainable transition of the concrete industry, a major contributor to global CO₂ emissions, demands both operational optimization and scientific innovation. This thesis presents comprehensive data mining frameworks for both industrial and literature source data to support the development of more sustainable concrete materials. Focusing on concrete manufacturing, we develop AI-powered methodologies tailored to real-world industrial data and complex scientific literature. For industrial data mining, we propose to incorporate interpretability and realistic engineering design scenarios to enhance the reliability of both predictive and prescriptive modeling of concrete mixes containing supplementary cementitious materials (SCMs). A domain-informed amortized Gaussian process and a shallow multi-layer perceptron (MLP) are shown to possess superior scientific consistency in predicting time-varied compressive strength, and time-invariant slump and air content properties, respectively. The explainable surrogate property models are applied in mix design optimization under a variety of realistic scenarios considering different engineering design requirements and SCM costs and densities. The importance of the comprehensive property constraint set is demonstrated in comparison against a baseline using only 28-day strength constraint which results in unreasonable property values. The necessity to differentiate realistic scenarios is also highlighted through the differences of optimized mixes and their production costs and climate impacts. Higher design strength, higher design slump, lower design air content, higher SCM density, and higher SCM unit cost can drive up the production costs. Though stratification patterns in the production costs of optimized mixes are observed across different scenarios, the mix-wise climate impacts are not clearly stratified, indicating that substantial emission reduction can be achieved without significantly increasing costs, regardless of the realistic scenarios. For literature mining, a novel method that finetunes lightweight large language models (LLMs) (pythia-2.8B) with multichoice instructions is developed. With the multifaceted linguistic complexity of communication within the domain rendering it infeasible to adopt the conventional named-entity-recognition approach, the new method successfully achieves great information inference accuracy in a time-, cost-, and computation-efficient manner, outperforming the GPT-3.5 in-context learning baseline by over 20%. A knowledge graph is constructed with the literature-mined data, offering insights to promote alternative material substitution strategies in concrete production as the current commercial SCMs are not comprehensively sustainable in the longer term. Statistical summary and temporal trend analyses are adopted to provide both static and dynamic insights into the research landscape. Although SCMs have remained a research hotspot, results revealed a systematic shift in recent studies from commercial SCMs to other materials. Geopolymer and fine aggregate studies have surged in the recent period, while clinker feedstock and filler studies have declined. A node similarity metric is modified to develop a model-free link prediction algorithm, enhanced with random graph perturbation for robustness and uncertainty quantification. Through link prediction, the currently underexplored lime-pozzolan cement application emerges as a potentially promising future research direction.

Metabolic Scaling Analysis of Building Energy Efficiency: A Case Study of Massachusetts Institute of Technology

Thu, 01 May 2025 00:00:00 GMT

Metabolic Scaling Analysis of Building Energy Efficiency: A Case Study of Massachusetts Institute of Technology Hsu, Yu-Hsuan The building sector plays a critical role in global energy consumption and carbon emissions, accounting for 21% of global GHG emissions (12 GtCO₂-eq) and 31% of global final energy demand (128.8 EJ) in 2019 (Cabeza et al. 2022). This reality underscores the urgent need to enhance energy efficiency within the sector. This research applies ecological metabolic scaling principles to building energy analysis, utilizing the Massachusetts Institute of Technology (MIT) campus as a case study. Analogous to biological systems, where an animal’s metabolic rate scales to 3/4 power of its mass, our findings indicate that larger buildings, similar to larger organisms, are inherently more energy efficient. Furthermore, an analysis of overall energy consumption at MIT from 2009 to 2020 reveals a steady decline, though not proportionally, as the scaling exponent fluctuated with a decreasing trend (<3/4), indicating improved efficiency in larger buildings. However, the COVID-19 pandemic in 2020 acted as a major shock, disrupting this trend. This disruption was likely driven by operational and behavioral changes, including reduced occupancy, increased remote work, and adjustments to ventilation and heating systems to ensure health and safety. These shifts highlighted the system’s tendency to return to the baseline scaling exponent of 3/4, demonstrating regression to the mean and ultimately pushing efficiency back to its prior baseline level of 25%. Additionally, the study includes case analyses of specific buildings on the MIT campus to provide deeper insight into comparative energy performance. While several guidelines for energy systems have been proposed, certain limitations remain. Future research should focus on expanding the dataset to help validate the applicability of these findings to other contexts while also accounting for variations in building types. Ultimately, this study aims to facilitate the development of more effective policies and innovations in building energy management.

High Strain-Rate Responses of Mechanical Metamaterials

Thu, 01 May 2025 00:00:00 GMT

High Strain-Rate Responses of Mechanical Metamaterials DesRoberts, Collin G. Mechanical metamaterials—materials with deterministic microstructures that attain unique combinations of properties—have revolutionized the parameter space of engineering materials over the last decade. While their quasi-static mechanical responses have been thoroughly characterized, their responses in the dynamic regime are not fully understood, especially at strain rates above 10^3 s^−1. Using microscale uniaxial compression and custom microscale Kolsky bar capabilities, we uncover the strain rate dependence of mechanical metamaterials over eight orders of magnitude, ranging from strain rates of 10^−3 to 10^5 s^−1. Herein, we describe the development and execution of direct impact experiments using a custom-built micro Kolsky bar set-up, delving into the details of its design, fabrication, and data analysis. We first characterize the rate dependence of the polymer used for sample fabrication, IP-S, and relate it to the responses of different metamaterial morphologies at the same strain rates. The results of these experiments uncover how geometry greatly affects rate dependence of mechanical properties in the dynamic regime. Understanding the high strain rate behavior of metamaterials is necessary to ensure reliable performance in real-world applications.

Scaling Carbon-Cement Supercapacitors for Energy Storage Use-Cases

Thu, 01 May 2025 00:00:00 GMT

Scaling Carbon-Cement Supercapacitors for Energy Storage Use-Cases Grewal, Darshdeep The urgent global transition to renewable energy is constrained by the intermittent nature of solar and wind sources, highlighting the critical need for scalable energy storage solutions. This thesis presents a comprehensive investigation into the development of structurally integrated supercapacitors based on carbon-doped cement composites, known as EC3 cells. These multifunctional materials combine structural performance with electrochemical energy storage capabilities, enabling integration directly into civil infrastructure. The research focuses on three essential challenges for real-world deployment: (1) replacing laboratory acrylic casings with hydrophobic sealants compatible with cementitious systems, (2) quantifying and mitigating shrinkage and swelling in nanocarbon cement matrices under electrolyte exposure, and (3) identifying corrosion-resistant current collectors that maintain conductivity and mechanical durability under harsh conditions. Bitumen-based coatings were found to be promising sealants for moisture containment. Shrinkage studies [ are underway, I will complete this part shortly]. Meanwhile, corrosion testing of various collector materials revealed that graphene sheets and stainless steel–reinforced graphillic papers offered optimal trade-offs between conductivity, corrosion resistance, and mechanical performance. The thesis concludes with two field-implementation design proposals—a vertical column and a vaulted arch—both of which leverage compression to improve electrochemical contact and stability. Altogether, this work establishes a foundational framework for embedding energy storage directly into the built environment.

Feedback Controller Design For Low-Power Autonomous Blimp Robot

Thu, 01 May 2025 00:00:00 GMT

Feedback Controller Design For Low-Power Autonomous Blimp Robot Ilerbaig-Bajona, Pau J. Low-power autonomous robots are able to do things that humans cannot do alone, ranging from small robots that traverse through the human body or other tight spaces to surveillance and monitoring robots that perform extended missions. Navigation is a key power draw for these robots, and thus, their motion planning must be designed to minimize power usage. In order to test motion planning algorithms, a low-power autonomous blimp robot that uses buoyancy to reduce power requirements for three-dimensional movement was designed. The blimp has two forward-facing rotors and two vertically-facing rotors. Due to the underactuated nature of the robot, the blimp cannot be translated sideways, thus leading to a coupling of the blimp’s two horizontal degrees of freedom. To control this, we present a controller for the blimp robot with three separate PID control loops: one for altitude control, one for angle control, and one for proximity approach. Additional fuzzy logic is implemented to improve performance and limit inefficiencies in the dynamic system and controller, such as turning towards the goal first before approaching forward. Combining the PID control loops and fuzzy logic allows for movement from a start point to a goal point, remaining within a 0.3 m radius of the goal point once it is reached. Further work that can be done to improve the physical system and controller is discussed, such as balancing the blimp gondola and rotors, as well as implementing different, physics-based controllers.

Preserving Human Autonomy in AI-Mediated Negotiations

Thu, 01 May 2025 00:00:00 GMT

Preserving Human Autonomy in AI-Mediated Negotiations Chen, J. Alvin The rapid integration of generative artificial intelligence (AI) into negotiation and conflict resolution processes raises critical ethical concerns about the erosion of human autonomy, particularly when AI systems navigate irreconcilable “sacred” values (non-negotiable moral principles) alongside transactional “mundane” interests. This thesis investigates whether generative AI can be designed to recognize and respect important values and beliefs while preserving human agency in decision-making. Drawing on datasets from a repository of large language model (LLM) prompts tested in simulated negotiation scenarios, this study employs a mixed-methods approach to evaluating AI’s efficacy in balancing efficiency with ethical imperatives in negotiation. Quantitative metrics (enumerating the outcomes of two-party negotiations) are analyzed alongside qualitative assessments of values such as transparency and consent, drawn from Kantian ethical frameworks. My analysis reveals that while AI negotiating bots excel in trades across mundane, tradable interests they struggle to navigate beliefs and values without oversimplifying moral reasoning or obscuring cultural considerations. These findings inform policy recommendations, including a call for human-in-the-loop validation and technical safeguards for protecting important values in efforts to incorporate AI-assistance into negotiations. By bridging technical analysis and ethical theory, I hope this research contributes to improvements in designing autonomy-preserving AI systems for use in a range of negotiating settings, prioritizing human dignity alongside computational efficiency.

Junctions and Strands: Breaking Property Tradeoffs in Polymer Networks and Composite Polymer Electrolytes

Thu, 01 May 2025 00:00:00 GMT

Junctions and Strands: Breaking Property Tradeoffs in Polymer Networks and Composite Polymer Electrolytes Herzog-Arbeitman, Abraham This dissertation first examines the mechanics of polymer networks, specifically material toughness and the nature of material fracture. Polymer networks, which include tire rubber, windmill turbine blades, tissue engineering scaffolds, polymer electrolytes (vide infra) and many other materials, possess a useful lifetime typically limited by a fracture event. Thus, methods of controlling toughness (the resistance of a material to tearing) without compromising composition or other properties would dramatically affect waste generation and energy use in the myriad applications in which polymer networks are employed. Toughness in rubbery polymer networks derives from the length and density of the polymer strands; thus, it is generally inversely related to stiffness, as captured in the classic Lake-Thomas theory. This inverse relationship has been perturbed through incorporation of forceresponsive molecules (mechanophores) that may either toughen or weaken the material depending on network construction and topology. The first part of this thesis identifies a new class of mechanophores called tetrafunctional cyclobutanes (TCBs), which can be used to either toughen or weaken a network of single topology without substantial change of network composition, even in dilute gels which are difficult to toughen by other methods. TCBs are then used to identify the mechanisms of mechanophore toughening or weakening in other networks, through a proposed topological metric called network strand continuity (NSC). We show that TCB substituents control the regio- and chemo-selectivity of the cyclobutane core under stress, and this molecular-level selectivity is responsible for network toughening or weakening on the macroscale. These effects can be predicted based on knowledge of activation energetics of the junction guided by NSC. Subsequently, effects of other network structure parameters on the magnitude of toughening or weakening are considered and the molecular design of second-generation highly active TCBs is described. The second part of this dissertation concerns the design of microporous polymer electrolytes and the applications of their composites and gels in batteries. Polymer electrolytes are a highly anticipated alternative to the liquid electrolytes currently in use, which are toxic, flammable, and incompatible with next-generation battery chemistries. Previous polymer electrolytes exhibit inadequate conductivity and a severe tradeoff between conductivity and mechanical properties. These challenges are accentuated in single-ion conductors, which are theorized to have the strongest rate capability. A new class of single-ion conducting polymer electrolytes that mimics the conduction mechanism of ceramic electrolytes to achieve strong mechanical properties, high conductivity, processability, stability, and recyclability is described. These polymers constitute the first regular microporous polyanions, and the most dissociative microporous polyanions to date. These polymers, alongside other rigid (but not microporous) polysulfonimides enable strong conductivity performance when coupled with suitable dopant (here succinonitrile) in low weight fractions. Low molecular weight controls flexible polymer analogs show inferior mechanical and conductivity properties. In fact, microporous composites outperform even liquid analogs. These composites show best-in-class combinations of mechanical and conductivity properties, and can even conduct divalent cations like Zn(II), a challenging but energy-dense battery metal. Simulations show that polymer-succinonitrile interactions enable fast conduction at the pore edge which results in synergistic behavior.

Testing Soccer Cleats Designed to Reduce ACL Tears

Thu, 01 May 2025 00:00:00 GMT

Testing Soccer Cleats Designed to Reduce ACL Tears Sandell, Remi One of the most common lower extremity injuries in soccer are ACL tears which have a long recovery time of 6-13 months and have a high risk of reinjury. The majority of ACL tears are non-contact and the mechanism behind them has been explained to be related to landing flatfooted. This provides an opportunity for engineers to design cleats to reduce the risk of these injuries. The company HBN shoes designed a soccer cleat aimed at reducing ACL tears by increasing the flexibility of the cleat around the metatarsophalangeal joint. The idea behind this is that increased flexibility at this joint will decrease landing flat-footed while running, reducing injury risk. This study evaluated the ability of the shoe to decrease flat-foot running and flatfooted landing while passing the ball using Fscan pressure sensors and GoPro cameras. For the majority of the participants in the study, the HBN cleats had a reduced peak heel force during running compared to the test cleats. When passing the ball wearing the HBN cleats, the majority of participants had a lower percentage of the planting step in a flat-footed position compared to the control cleats. This indicates that the HBN cleats could be effective in reducing flat-footed running and landing in athletes.

Race Car Reverse Gear Design

Thu, 01 May 2025 00:00:00 GMT

Race Car Reverse Gear Design English, Ashley E. This thesis presents the design, simulation, and installation of a reverse gear system for the RUSH SR, a lightweight, motorcycle-engine-powered race car that lacks built-in reverse capability. The proposed solution repurposes a high-torque automotive starter motor to drive the car in reverse through engagement with a custom ring gear on the rear differential. Analytical modeling and time-domain simulation were used to evaluate performance under varying loads, including the effect of incline angle on terminal velocity and motor current draw. Simulated results show that the system can reliably move the car in reverse on slopes up to 10° before stalling, with current draw remaining within safe operational limits. The mechanical design includes a new differential carrier, gear coupler, and ring gear, while the electrical system explores both off-the-shelf and custom high-side switching controllers to manage power and solenoid activation. The final hardware was bench tested and installed on a working vehicle. Recommendations for future validation include current-limited incline testing and dynamic vehicle response trials. This modular and cost-effective system demonstrates a practical solution to a common race car limitation while preserving the RUSH SR’s lightweight performance characteristics.

Design and Characterization of a Pressure Controller Test Stand for Soft Pneumatic Actuator Testing

Thu, 01 May 2025 00:00:00 GMT

Design and Characterization of a Pressure Controller Test Stand for Soft Pneumatic Actuator Testing Comiskey, Evan L. Soft pneumatic actuator design has been of interest to the soft robotics community in recent years for their potential to expand the roles of robotics systems in human-interacting medical devices and delicate manufacturing processes. Unfortunately, it is difficult to analytically predict the degradation behavior over time of soft pneumatic actuators, and to compare such behavior in a repeatable manner between different actuator designs. To gather such degradation data experimentally, this thesis presents an open-source proportional-integralcontrol (PID) control test stand system with only off-the-shelf components, in order to create a tool for standardized and repeatable characterization testing of soft pneumatic actuator life-cycle behavior. The system is open-source and modular, allowing researchers in the MIT Fabrication-Integrated Design Lab (FIDL) as well as other soft robotics researchers to customize the program and associated hardware as necessary for their own soft pneumatic actuator research inquiries. This thesis explores the design of both the hardware and software of this test stand system, which is informed by the system’s functional requirements, and additionally explores the PID controller’s capabilities and limitations.

Characterizing Effective System Architectures for Cislunar Space Situational Awareness

Thu, 01 May 2025 00:00:00 GMT

Characterizing Effective System Architectures for Cislunar Space Situational Awareness Rude, Connor D. Achieving Space Situational Awareness (SSA) in the Cislunar region—the area between the geosynchronous belt and the Moon's gravitational boundary—poses significant technological and organizational challenges. Instead of proposing new theoretical systems, this thesis employs the Architecting Innovative Enterprise Strategy (ARIES) Framework to evaluate existing SSA architectures and previously suggested solutions. ARIES provides a structured assessment through its elements (strategy, information, infrastructure, products, services, processes, organizations, and knowledge), identifying infrastructure, acquisition strategies, policy-driven timelines, and communication structures as key areas for improvement. Stakeholder objectives, current initiatives, and operational needs guide the characterization of an ideal SSA architecture. Four prior system proposals for cislunar SSA are assessed using qualitative analysis of existing literature and first-order physics-based simulations. These evaluations correlate specific design features with enhanced system suitability. Particularly beneficial are constellation proximity to targets, strategic constellation placement and phasing, sensor orbital diversity, and orbital stability. Additionally, certain design strategies consistently yield higher suitability, including focusing on underserved SSA regions, leveraging heritage technology, and optimizing designs for ride-share launch compatibility.

Exploring Visual and Haptic Feedback Systems on User Performance with a Hand-Held Robot

Thu, 01 May 2025 00:00:00 GMT

Exploring Visual and Haptic Feedback Systems on User Performance with a Hand-Held Robot Shiferaw, Ruth While robotic systems allow users to maintain accuracy in high-precision environments, achieving intuitive control is challenging without real-time feedback. Haptic feedback, which communicates otherwise unfelt sensations through vibrations, is widely used in consumer technologies such as video games and smartphones. However, in contexts where knowing the precise force applied by the robot is critical—such as medical procedures or hazardous environments—haptic cues alone may provide insufficient resolution, increasing user workload. Visual feedback, by contrast, is more commonly used and offers greater versatility and precision. This study compared the impact of visual feedback (a color-changing LED light strip) and haptic feedback (vibrations in a controller) on user performance in a “fragile object” manipulation task. Nine participants completed the task under four feedback conditions: no feedback, visual feedback, haptic feedback, and combined visual-haptic feedback. Subjective ratings showed that most participants preferred modalities that included visual cues, citing lower perceived workload and clearer force awareness. However, despite some participants reporting minimal benefit from haptics, performance metrics revealed that for others, haptic feedback meaningfully supported task success. These findings suggest that while simple visual indicators, such as green-yellow-orange-red light strips, provide accessible and interpretable force feedback, the integration of haptic cues can enhance performance by offering complementary real-time force information. Future designs may benefit from refining both modalities to balance intuitiveness, resolution, and user comfort, especially in applications requiring precise force modulation.

LLM-Supported Natural Language to Bash Translation

Thu, 01 May 2025 00:00:00 GMT

LLM-Supported Natural Language to Bash Translation Westenfelder, Finnian Ellis The Bourne-Again Shell (Bash) command-line interface for Linux systems has complex syntax and requires extensive specialized knowledge. Using the natural language to Bash command (NL2SH) translation capabilities of large language models (LLMs) for command composition alleviates these issues. However, the NL2SH performance of LLMs is difficult to assess due to inaccurate test data and unreliable heuristics for determining the functional equivalence of Bash commands. We present a manually verified test dataset of 600 instruction-command pairs and a training dataset of 40,939 pairs, increasing the size of previous datasets by 441% and 135%, respectively. Further, we present a novel functional equivalence heuristic that combines command execution with LLM evaluation of command outputs. Our heuristic can determine the functional equivalence of two Bash commands with 95% confidence, a 16% increase over previous heuristics. Evaluation of popular LLMs using our test dataset and heuristic demonstrates that parsing, in-context learning, in-weight learning, and constrained decoding can improve NL2SH accuracy by up to 32%. Additionally, we consider military use cases for NL2SH models and discuss the limitations of current Department of Defense documentation standards for LLMs. We write and publish documentation for our models and datasets to promote safe use. Our findings emphasize the importance of dataset quality, execution-based evaluation, translation method, and proper documentation for advancing NL2SH translation and enabling responsible use. Our code is available at https://github.com/westenfelder/NL2SH.

Abiotic and Biotic Polymer Degradation to Inform Sustainable Design

Thu, 01 May 2025 00:00:00 GMT

Abiotic and Biotic Polymer Degradation to Inform Sustainable Design Tantawi, Omar As global plastic production continues to rise, understanding environmental processes governing plastic degradation is crucial to inform the sustainable design of polymers. This thesis is structured into three chapters, each addressing critical aspects of polymer degradation: In the first chapter, I develop and apply a sequential abiotic (photodegradation and hydrolysis) and biotic degradation test to a diverse suite of 18 polymers, including novel polyhydroxyalkanoates polyesters, commercially available bio-based polymers (e.g., polylactic acid, poly-3-hydroxybutyrate), and conventional fossil-derived polymers (e.g., polypropylene, polyethylene terephthalate). Results illustrate that current biodegradation standard methods relying only on mineralization underestimate polymer degradation by up to two-fold. Simulated sunlight notably enhanced polymer degradation by mobilizing dissolved organic carbon (DOC), which proved highly biodegradable in marine environment. Chemical structural differences were clearly linked to degradation behaviors, emphasizing the utility of the developed workflow for rapidly identifying environmentally relevant degradation mechanisms, which can inform structure-property relationships for future polymer designs. In the second chapter, I delve deeper into characterizing polymer-derived dissolved degradation products. Conducting Mass Remainder Analysis (MARA) using non-target liquid chromatography–high-resolution mass spectrometry (LC-HRMS) data, we systematically identified oligomeric degradation products and homologous series of polyamide-6 (PA6), polycaprolactone (PCL), and polylactic acid (PLA). Complementary experimental approaches (retention-time shifts across varied mobile phase pH, fragmentation analysis, and spectral matching) were essential to improve structure elucidation and determine acid-base properties (pKa) and hydrophobicity (logKow and logD). The experimental findings emphasized large deviations of oligomers hydrophobicity from computational predictions, underscoring the necessity for oligomer-specific experimental data to enhance environmental fate modeling and risk assessment accuracy. In the third chapter, I investigate the fate of polymer-derived dissolved organic carbon (p-DOC) from PLA, PCL and PA6, focusing specifically on oligomer chemistry. Using natural marine microbial communities, PLA- and PCL-derived DOC demonstrated rapid biodegradation (82-85% within six days), while PA6-derived DOC exhibited resistance. Detailed analysis using high-resolution mass spectrometry and MARA revealed significant chemical structure dependence in biodegradation rates, with rapid degradation of aliphatic ester-containing cyclic and linear oligomers. Larger cyclic oligomers degraded faster, while short linear oligomers showed transient accumulation followed by degradation. PA6 oligomers exhibited limited biodegradability, with cyclic oligomers showing minimal degradation. The results emphasize the critical influence of oligomer chemistry and microbial enzymatic specificity, providing essential insights for designing sustainable polymers compatible with marine environments.

Simulation-Based Reinforcement Learning Policy Optimization for Tactile Manipulation: A Case Study on the Eyesight Hand

Thu, 01 May 2025 00:00:00 GMT

Simulation-Based Reinforcement Learning Policy Optimization for Tactile Manipulation: A Case Study on the Eyesight Hand Chang, Ethan Robotic manipulation remains a complex and unsolved challenge due to the need for adaptability across diverse objects and tasks. In this work, we explore how to train effective manipulation policies using reinforcement learning in simulation for the Eyesight Hand: a novel, low-cost, tactile-enabled robotic hand. We implement a range of experiments in MuJoCo to evaluate the impact of controller types, observation spaces, reward formulations, and curriculum strategies on policy performance. Our findings highlight the benefits of delta position control, a carefully selected observation space including joint states, control vectors, object pose, and contact forces, and success-driven curriculum learning. Our study establishes baseline strategies for training robust, tactile-based policies on this in-house hardware.

Designer-User Validation of a Method for Predicting Cognitive Interventions for Sustainable Product Design

Thu, 01 May 2025 00:00:00 GMT

Designer-User Validation of a Method for Predicting Cognitive Interventions for Sustainable Product Design Ladolcetta, Mia Life cycle assessments to measure the environmental impact of products is often limited by a lack of information about the use phase of a product. To aid this gap by encouraging sustainable user behavior from the early brainstorming and prototype phases of product development, principles of environmental psychology and sustainable product design can be linked through determinants of pro-environmental behavior and cognitive interventions included as product features. A study will be completed to validate these relationships by surveying participants on their determinants, grouping participants into personas that share similar determinant ratings, then pitching sketch prototypes developed with corresponding cognitive interventions and measuring their receptiveness to each sketch prototype. To aid in the development of this sketch prototype evaluation survey, a series of user interviews were conducted with the sample use case of factors that encourage the use of reusable water bottles to better understand how users may evaluate the products depicted in the sketch prototypes. Validating the links between determinants and cognitive interventions can assist designers in determining a hierarchy in the determinants they design for and more efficiently select effective cognitive interventions for users with different experiences.

Hyperspectral Remote Sensing for UXO Detection and Damage Assessment on Airfield Pavements

Thu, 01 May 2025 00:00:00 GMT

Hyperspectral Remote Sensing for UXO Detection and Damage Assessment on Airfield Pavements Pietersen, Randall If an airfield being operated by the U.S. Air Force is attacked, the current method for assessing its condition is a slow visual and manual inspection process, exposing personnel to dangerous conditions and delaying repair operations. Developing a fully autonomous remote assessment solution would improve the speed and safety of this critical task, but remains an unsolved problem despite continued advances in drone technology, deep learning, and computer vision. This research explores using near-surface hyperspectral sensors as an alternative to red, green, blue (RGB) digital cameras, in hopes of improving detection precision and accuracy for airfield assessment. However, even with modern hyperspectral sensors the benefit of increasing spectral image resolution comes at a cost, creating addition complexity, uncertainty, and sensitivity in the acquisition, data correction, and downstream detection processes. This work presents a series of tests, each designed to better understand and refine a full hyperspectral image detection sequence, starting with sensor selection and raw data acquisition, proceeding to radiometric correction, and culminating in image recognition by means of supervised deep learning (DL). Regarding sensor selection and data acquisition, these findings indicate that for many applications of computer vision, using a hyperspectral camera with high spectral resolution is unnecessary. It is more beneficial to select a camera with snapshot imaging that instead maximizes spectral range or spatial resolution. Radiometric correction is then explored, and experiments demonstrate that correction makes machine learning classification models less sensitive to changes in scene illumination, thus improving overall image recognition performance. Finally, deep learning models for image recognition are tested and a new method for generating synthetic hyperspectral data is developed and shown to be useful for estimating hyperspectral model performance on larger datasets, when real data are limited. Overall, the findings presented in this thesis suggest that by refining the methods used for data acquisition, correction, and detection, hyperspectral imaging improves image recognition when compared to traditional RGB cameras. This applies not only for airfield damage assessment but extends to other real-world applications requiring computer vision and scene understanding.

New Measurement Approaches to the Study of Secondary Organic Aerosol

Thu, 01 May 2025 00:00:00 GMT

New Measurement Approaches to the Study of Secondary Organic Aerosol Helstrom, Erik Aerosol particles constitute a class of atmospheric pollutants that are detrimental to human health and influence the Earth’s climate. A significant fraction of aerosol mass is composed of organic material, produced by photochemical reactions of organic trace gases which form secondary organic aerosol (SOA). However, the large diversity of volatile organic compounds (VOCs) makes it challenging to identify all of the chemical reactions contributing to SOA formation. In addition to this chemical complexity, our ability to identify and measure all of the relevant organic compounds, especially species present in aerosol particles, is limited by challenges in efficiently sampling and detecting the various classes of molecules formed. Improving knowledge of the chemical behavior of aerosol will improve our ability to predict how changing emissions and chemical conditions will impact the formation and properties of particulate matter in the future. This thesis will explore recent improvements in instrumentation and measurement techniques and apply them to laboratory studies of organic carbon and SOA. First, we adapt a technique for measuring total suspended carbon to laboratory chamber experiments, converting organic compounds with high temperature catalysis to carbon dioxide, which is then monitored in real time. This allows for a “top-down” constraint on the overall concentration of all organic species (including SOA) as experiments proceed, as some lower volatility products are lost to the surfaces of the laboratory chamber. Second, we compare the measurements of SOA from three chemical ionization mass spectrometers using different ionization and desorption methods to detect particle-phase species. Clear differences emerge in the detected formulas across instruments, highlighting variations in chemical sensitivities to different classes of compounds and the influence of fragmentation on the detected products. Finally, we explore how changing peroxy radical fate influences SOA formation by monitoring SOA composition with extractive electrospray ionization (EESI) mass spectrometry. Differences in particle-phase products, particularly nitrates, hydroperoxides, and dimers, make the dependence of initial SOA composition on peroxy radical pathways clear. Over time, we observe a convergence of SOA spectra formed under different peroxy radical regimes, suggesting the influence of secondary products and particle-phase chemistry, though some differences persist from the initial gas-phase peroxy radical fate. Overall, this thesis demonstrates improved tools for constraining and investigating VOC oxidation pathways leading to particle-phase organic species.

Effect of Recycling Stream Contaminants on Extrusion and Mechanical Properties of rPET

Thu, 01 May 2025 00:00:00 GMT

Effect of Recycling Stream Contaminants on Extrusion and Mechanical Properties of rPET Escandon, Mercedes The emerging field of large-scale additive manufacturing has a wide range of applications including the potential to use recycled plastics. A type of 3D printing that can use recycled plastics is Fused Granulate Fabrication, which takes in polymer pellets and melts them down in a screw and barrel before extruding them layer by layer. The process of turning plastic trash into print-ready pellets typically involves sorting, shredding, washing, drying, and re-extruding the material into pellets, each step requiring time, equipment, and energy. To reduce both the cost and carbon footprint of 3D printing with recycled plastic, some of these steps could potentially be eliminated. Washing and drying plastic is a very energy-intensive step that could potentially be skipped or significantly modified. The possibility of using unwashed plastics in 3D printing was explored, focusing on residual beverage contamination. Controlled amounts of soda were introduced to clean, virgin PET pellets prior to drying. The contamination levels ranged from 0.25% to 3% soda by mass. The plastic was melted and extruded into thin strands using a stationary horizontal extruder and a conveyor belt. The mass flow rate was determined from the mass of strands to quantify the quality of extrusion. No significant effect of contamination level on mass flow rate was measured. The strength of the parts was determined using a tensile test. The Young’s Modulus initially increases with the contamination level, peaking at 1% contamination with a Young’s Modulus of 2.80 GPa, which was 30% higher than the measured value for the clean PET. Above 1% contamination, there was a significant drop off strength. These results demonstrate that there is an acceptable level of beverage contamination when recycling plastic.

Measuring Density of Novel Technorganic Material

Thu, 01 May 2025 00:00:00 GMT

Measuring Density of Novel Technorganic Material Ramos-Munoz, Jorge Felix Self-healing technorganic materials are critical for organ printing/repair operations. This study characterized one such technorganic material by measuring its density. By ascertaining the density of the material, we hope to later derive other important properties including: the speed of sound within the material, its stiffness, its conductivity, and its natural frequency. The material in question was printed on a gold substrate by depositing hexane within a silver nitrate solution. The density was obtained by observing the volume change of the solution in which it was printed and by measuring the sample’s mass. A final density of 0.622±pm 0.227 g/cm³ was measured, and we hope to continue characterizing the material’s mechanical, thermal, and electrical properties with future studies.

Design and Testing of a Hadal Sediment Sampling System

Thu, 01 May 2025 00:00:00 GMT

Design and Testing of a Hadal Sediment Sampling System Do, Thao X. Sediment cores are conventionally gathered to collect data on seabed chemical and mineral composition. These samples are important in determining the health of the surrounding environment and, if sediment layers are preserved, assessing the environmental history and trajectory of our ocean floors. Over the last few decades, some attempts to collect sediment core samples have been made from as deep as the Mariana Trench, which contains the deepest known point on Earth’s surface at 11,000 meters [1] [2]. However, the hadal region remains one of the most underexplored areas of our oceans [3]. Working in conjunction with Inkfish, a submersible technology company, we developed a deep-sea sediment core sampler that will travel to the Mariana Trench aboard one of Inkfish’s submersible landers and will collect four inches of sediment in ambient pressures of 110.32 MPa [4]. The first phase of this project was to design and fabricate a prototype sediment core sampling device. We designed an entirely mechanical sampler because underwater actuators suitable for use in hadal conditions are difficult to source and require additional communication and power resources. Furthermore, our device aims to preserve the layers of the sediment core samples as the lander ascends to preserve relative time scales within the sample. The next phase involved testing the functionality of different subsystems of our device. In this paper, we considered two different sediment collection apparatuses and one-way valves for our collection tube. We performed field testing at the Charles River in Cambridge, MA to assess which apparatus and valve combination would provide the best results based on the volume of sediment collected and retained. Building on our mechanical proof-of-concept, we improved the design and fabricated a second iteration prototype for 2-3 km depths. In this version, we addressed the two main weaknesses of the initial proof of concept design: substantial friction during sliding and the complexity of assembly and maintenance due to the numerous parts. Lastly, we engineered a deployment system that made our device compatible with Inkfish’s lander deployment procedure. The deployment system was designed to lower the sediment sampler device from within the bay to below the lander once the lander was submerged in water. In early October, the improved sediment sampler and the deployment system were tested during an engineering expedition near the shores of Tonga. Ultimately, we believe that our sediment sampler presents a viable purely mechanical solution to collecting deep-sea sediment from profoundly unexplored areas at hadal depths like the Mariana Trench. Our sampler can easily be mounted onto any surface where it would touch the ocean floor, requiring no electronics or controls. Though we were constrained by the particular seafloor lander used by Inkfish, the size of the sampler is scalable, allowing both the sample diameter and depth to be adjusted for a given mission. By making these sediment samples more accessible, we believe we will have an impact across a number of marine research areas.

Redefining “Space Sustainability” for Launch Vehicles: Forecasting the Atmospheric Impact of the Commercial Space Launch Industry in 2050

Thu, 01 May 2025 00:00:00 GMT

Redefining “Space Sustainability” for Launch Vehicles: Forecasting the Atmospheric Impact of the Commercial Space Launch Industry in 2050 Ma, Clara Z. Discussions on “space sustainability” have largely centered on orbital debris, the burnup of vehicles during atmospheric reentry, and the resulting emissions. However, few studies have examined emissions from the launches themselves. Along with reentry burnup, rocket launches are the only source of high altitude anthropogenic emissions. At such high altitudes, emitted particles can remain in circulation for years. With the annual growth rate of the commercial launch industry averaging 14.6% in the last 4 years and over 211 launches in 2023 alone, our research on the atmospheric impact of launch vehicles comes at a crucial point in the policy debate on space sustainability. This thesis outlines several potential future scenarios of the launch industry in 2050, with all the vehicles in each scenario using the same fuel type. We examine these four launch scenarios—a kerosene (RP-1) launch industry, a methane (CH4) launch industry, a hydrogen (H2) launch industry, and a control or “baseline” scenario without launches. For each scenario, we estimate the number of launches for a distribution of heavy-lift launch vehicles across origin spaceports. We simulate the chemical interactions of the launch plumes with the atmosphere using the global atmospheric chemistry model GEOS-Chem High Performance (GCHP). Finally, we quantify the steady state impact of launch emissions on stratospheric ozone and surface air quality. We find that the black carbon emitted by kerosene and methane rockets causes an indirect increase in stratospheric ozone due to the removal of NOx, with ozone column change averaging 5.07 Dobson Units (DU) and 1.26 DU respectively; hydrogen rockets cause a net decrease in ozone column averaging -0.11 DU. The population-weighted average surface ozone impact is -0.286 ppb, -0.068 ppb, and 0.023 ppb for RP-1 rockets, CH4 rockets, and H2 rockets respectively. The population-weighted average surface PM2.5 impact is -0.031 μg/m3, -0.004 μg/m3, and 0.002 μg/m3 for RP-1, CH4, and H2 rockets respectively. Although RP-1 and CH4 rockets decrease surface ozone and surface PM2.5, H2 rockets have the smallest magnitude impacts on the atmosphere overall. Our findings have important implications for commercial launch providers, research institutions, and policymakers including the Federal Aviation Administration (FAA) and NASA.

Engineering Principles for Scalable Connectomics

Thu, 01 May 2025 00:00:00 GMT

Engineering Principles for Scalable Connectomics Garzon Navarro, Monserrate Brain tissue sectioning presents a significant challenge in connectomics, particularly when scaling to larger volumes. In the MICrONS 1 mm³ mouse visual cortex dataset, 25.1% of scanned images—representing over a month of imaging work—were discarded due to sectioning defects. Current methods result in material loss during cutting and face limitations in tool wear and process efficiency. This thesis examines tissue sectioning through an engineering lens. Drawing from established machining practices and parallel industries, we propose and evaluate potential improvements to sectioning methods. The work aims to contribute to ongoing efforts in mapping larger connectomes, making it more practical and less error-prone.

AI Trust and Technology Optimism in the Workforce: Data-Driven Insights into Regional Variation

Thu, 01 May 2025 00:00:00 GMT

AI Trust and Technology Optimism in the Workforce: Data-Driven Insights into Regional Variation Velonia Bellonia, Maria Eleni Automation and AI systems are reshaping the workplace. How these technologies make a difference varies according to local contexts. Workers’ willingness to trust and embrace these technologies is shaping how this transformation unfolds in practice. Some workers trust AI more than others, and interestingly, trust levels differ from one region to another. Drawing on a far-reaching 2024 worker survey spanning different countries, and on a rich body of literature on technology, trust, and change, this work examines how key factors influencing workers’ AI trust and technology optimism interweave, shaping their perspectives on new technologies and automation. The focus is on understanding how the industrial and regulatory landscape in which workers operate, combined with their personal experiences with AI, shapes their AI optimism, with a particular emphasis on the US and Europe. While external market innovation indicators provide limited understanding of workers’ technology optimism, individual interaction and familiarity with AI, alongside organizational AI adoption and a worker’s industry of employment, emerge as key factors shaping AI trust. Additionally, the regulatory environment, encompassing technology governance, social safety nets, and workers’ institutional trust, all seem connected with how workers think about the impact of new technologies on society, the economy, and their jobs. Interpersonal trust propensity contributes to AI trust formation, though its relevance exhibits regional variation. By offering insights into the critical factors shaping the relationship between workers and AI, this study aims to provide evidence that supports societies in unlocking the value of emerging technologies, while empowering the workforce to confidently embrace and excel alongside them.

Ideology and the First Gothics

Mon, 01 Jan 1990 00:00:00 GMT

Ideology and the First Gothics Grove, Allen Whitlock Thesis: B.S., Massachusetts Institute of Technology, Literature

Computational methods for dissecting multicellular mechanisms of complex diseases

Thu, 01 May 2025 00:00:00 GMT

Computational methods for dissecting multicellular mechanisms of complex diseases Mitchel, Jonathan Single-cell genomics technologies have enabled unbiased characterization of cell types and cellular states. However, the high-dimensional nature of this data necessitates computational and statistical methods to uncover the biological processes that shape it. In my thesis research, I developed three computational methods to explore genetic regulatory mechanisms underlying common diseases and the resulting multicellular patterns of dysfunction. In the first project, I developed a method called scITD to investigate how cellular processes across distinct cell types coordinate in disease contexts. scITD identifies sets of genes in one or more cell types that co-vary together across biological samples. Through the application of this tool to various immune-cell datasets, we uncovered highly reproducible gene expression patterns associated with autoimmune patient phenotypes. In the second project, I characterized technical artifacts prevalent in imaging-based spatial transcriptomics data. These artifacts arise from the misassignment of transcript molecules to incorrect cells. I further demonstrated how these artifacts confound downstream analyses, including differential expression and cell-cell interaction inference. To address this, I jointly developed a correction method that mitigates these artifacts, thereby uncovering novel biological insights in cancer datasets. In the third project, I introduced a computational method to unravel the mechanisms of genetic variants identified from genome-wide association study loci. This method tests whether these same genetic variants also underly changes to gene expression in specific cell types or states. Applying this tool to autoimmune and neurodegenerative datasets uncovered new SNP-gene-phenotype links and localized their effects to specific cell populations, helping to refine our understanding of these pathologies.

Scalable Packaging and Integration Solutions for Next-Generation Photonic Systems

Thu, 01 May 2025 00:00:00 GMT

Scalable Packaging and Integration Solutions for Next-Generation Photonic Systems Ranno, Luigi The ever-increasing demand for faster and more efficient computation has propelled the rapid growth of integrated photonics, with commercial products starting to reach global markets in recent years. Nevertheless, integrated photonics still lacks the scale required to meet market demands and falls short of the performance targets necessary for many critical applications. Innovative solutions are imperative if photonics is to drive technological advancement and become a ubiquitous part of next-generation systems, rather than being confined to niche or high-end applications. Among the key bottlenecks is photonics packaging, which refers to the challenge of electrically, optically, and thermally interfacing with a photonic integrated circuit (PIC). Current packaging solutions often impose significant design tradeoffs, contributing to industry fragmentation and high costs. Two-photon lithography (TPL), a high-resolution 3D manufacturing technique, has emerged as a promising enabler of robust and efficient optical interconnects. However, existing research has focused heavily on performance, often relying on additional chip processing steps (e.g., cladding removal) that hinder scalability. Moreover, prior work largely restricts itself to parameterized geometries, such as quadratic curves or spherical sections, that underutilize the true design freedom of TPL. My work addresses both of these limitations. I developed a freeform, facet-attached micro-reflector solution that is fully compatible with standard foundry processes, adaptable to challenging coupling scenarios, and computationally efficient to design. This coupling solution demonstrates all the properties desired in an ideal optical interface: low insertion loss (~0.6 dB), wide bandwidth (>300 nm), foundry compatibility, and geometric universality across PIC platforms. Another major challenge facing the photonics industry is the lack of critical functionalities within current foundry processes due to limited material availability. Significant gains in performance and capability can be realized by integrating new materials on-chip, but doing so while maintaining CMOS-foundry compatibility remains a formidable task. To address this, I helped develop a novel photonics platform enabling substrate-inverted multi-material integration. This platform supports seamless integration of diverse materials while leveraging existing PIC process stacks, including metallization layers, unlocking new classes of high-performance devices. Building on this idea, I further demonstrated how material integration can directly enable new applications. Specifically, I developed a selective and ultra-sensitive environmental lead (Pb²⁺) sensor, based on a crown ether functionalization layer. This device showcases the potential of hybrid material platforms to deliver practical, field-relevant solutions in environmental monitoring and beyond.

Synthetic and Post-Synthetic Methods towards Fine Tuning the Chemical and Physical Properties of Metal-Organic Frameworks

Thu, 01 May 2025 00:00:00 GMT

Synthetic and Post-Synthetic Methods towards Fine Tuning the Chemical and Physical Properties of Metal-Organic Frameworks Iliescu, Andrei This thesis explores synthetic and post-synthetic strategies for tailoring the chemical and physical properties of metal-organic frameworks (MOFs), with a particular emphasis on modulating redox activity, framework composition, and ionic conductivity. The first part of the work focuses on leveraging MOF-embedded polynuclear metal clusters for multi-electron redox chemistry. A square-planar tetramanganese cluster was shown to reversibly interconvert between molecular oxygen and metal-oxo species via a four-electron pathway. This reactivity was then investigated by varying the identity and redox potential of the metal centers within the tetrametal cluster. The Fe(II) and Co(II) analogs reveal distinct metal-specific behavior and provide insight into the tunability of redox-active SBUs within MOFs. Next, post-synthetic cation exchange was employed to access a previously unreported Zn-based MOF, ZnZnBTT, which exhibits significant Zn-ion conductivity due to mobile charge-balancing cations. This material demonstrates the potential of MOFs in next-generation solid-state battery technologies. Finally, the impact of linker electron donicity on cluster structure and reactivity was explored using a new mixed-azolate ligand. Four isostructural MOFs incorporating Co, Ni, Cu, and Cd were synthesized, revealing that the electron-rich pyrazolate groups modulate cluster composition and redox behavior. Notably, CoBTDP exhibits O₂ reactivity, unlike its all-tetrazolate counterpart, underscoring the role of linker design in tuning MOF function. Together, these studies demonstrate how careful control over MOF synthesis and post-synthetic modification can be used to fine-tune redox behavior, framework composition, and ion transport, providing new avenues for the design of functional porous materials.

Leveraging Metal Complexes for Optical Read-out of Magnetic Fields

Thu, 01 May 2025 00:00:00 GMT

Leveraging Metal Complexes for Optical Read-out of Magnetic Fields Yi, Seungyeon Optical detection of magnetic phenomena offers a compelling pathway toward the development of highly sensitive and versatile molecular sensors. This thesis investigates the design of metal complexes tailored for magnetic field read-out through light–matter interactions, focusing on two strategies. The first section explores magnetochiral dichroism (MChD), an optical effect that emerges from the interplay between molecular chirality and magnetism. By systematically varying the metal centers within a series of chiral lanthanide complexes—specifically, Tb³⁺ and Dy³⁺—we examine how differences in magnetic moment modulate the MChD response. This comparative study reveals fundamental chemical design principles for enhancing MChD intensity and deepens our understanding of how structural and electronic factors jointly shape this directional optical effect. The second section addresses the challenge of engineering optically addressable molecular qubits based on Ni²⁺ complexes. Realizing effective spin-state read-out in these systems requires precise control over both magnetic and photophysical properties. To this end, we investigate ligand modification strategies aimed at enhancing luminescence while preserving an S = 1 ground state suitable for quantum applications. Collectively, we hope these studies contribute to a better understanding of the design space for spin–photon coupled molecular systems, offering new tools for magnetooptical sensing.

Quantifying Plasticity and Temperature in High Velocity Microparticle Impacts

Thu, 01 May 2025 00:00:00 GMT

Quantifying Plasticity and Temperature in High Velocity Microparticle Impacts Lucas, Tyler J. The Laser Induced Particle Impact Test, or LIPIT, is a benchtop experimental setup that enables in-situ observation of micron-scale particles impacting targets at velocities ~10-1500 m/s. Through a combination of the high velocity and small length-scale of the impact, strain rates exceeding 10⁷ /s can be achieved while maintaining a subsonic plastic wave, preventing formation of strong shockwaves and hydrodynamic behavior. The LIPIT has been effectively applied to study phenomena in mechanical behavior, cold spray, and astronomical impacts, all of which will be further investigated in this work. This thesis combines LIPIT experiments and finite element modeling to explore the dynamic behavior of pure metals in the unique regime of strain, strain rate, and pressure achieved in high velocity impact conditions. First, the effect of material microstructure on the mechanical behavior of copper at high strain rates is explored to improve the capability of constitutive strength models in accurately representing experiments. Next, a method is introduced to measure the dynamic yield strength of ductile microparticles, effectively removing the need for the tacit assumption that the properties of bulk materials can be imposed on powders despite differences in processing. The understanding of microstructure and particle behavior are then combined to study the influence of material microstructure on the solid-state bonding of copper particles to copper substrates of different temper. Finally, this work applies the new understanding of plasticity and dynamic modeling in high strain rate conditions to quantitatively study the behavior of metals with a phase transition in absence of strong shockwaves.

Endothelial cell plasticity as a marker of vascular disease and predictor of adverse outcomes to stress

Thu, 01 May 2025 00:00:00 GMT

Endothelial cell plasticity as a marker of vascular disease and predictor of adverse outcomes to stress Salazar Martín, Antonio Gabino Endothelial cells (ECs) dynamically sense and adapt to their local biomechanical and biochemical environments, a process crucial for maintaining vascular homeostasis. Loss of this plasticity is implicated in vascular diseases, where endothelial dysfunction and maladaptive responses exacerbate disease progression and limit therapeutic efficacy. We investigated the role of endothelial cell plasticity under pathophysiological conditions and its impact on therapeutic interventions – mechanical, pharmacologic and genetic. Specifically, Aim 1 characterizes the modulation of EC plasticity by shear stress, revealing that flow patterns drive distinct transcriptional signatures and subpopulation behaviors, as demonstrated through singlecell transcriptomics in human aortic endothelial cells. Aim 2 examines the interplay between EC dynamism and antiproliferative drugs, in particular rapamycin and paclitaxel – the agents released from drug-eluting stents, showing that biomechanical cues from flow dominate EC responses, potentially limiting drug efficacy in regions of disturbed flow. Aim 3, extends the investigation by moving beyond the overwhelming of cells with pharmacologic dosing into the domain of controlled genetic modification, which is in concert with the direction of modern therapeutics and also provides a further dimension to the perspective of endothelial biology. We sought to discern if genetically modified cells maintain their characteristic "endothelial" profile or if the interplay among genomic alterations, transcriptional and proteomic shifts, and environmental cues leads to a state that challenges the hypothesis that ECs remain plastic until they become committed to flow. The integration of single-cell transcriptomics and in vivo models provides novel insights into the heterogeneity of endothelial responses and underscores the importance of considering biomechanical and biological factors in developing targeted therapeutic strategies for vascular diseases.

Predicting Ridership and Travel Time Impacts of Bus Service Changes Using Sketch Planning Methods

Thu, 01 May 2025 00:00:00 GMT

Predicting Ridership and Travel Time Impacts of Bus Service Changes Using Sketch Planning Methods Lim, Tiffany M. Bus service changes range in scale, and understanding their impacts on ridership and travel times can inform decision-making as changes are considered for the bus network. Budgetary limitations are at the heart of service change decisions, resulting in the need for analysts to assess different scenarios and accommodate quick turnarounds. This thesis provides a sketch planning framework for predicting ridership and travel time impacts of bus service changes, with a focus on direct demand models and the use of an open-source multimodal routing algorithm. The framework is designed to be streamlined with the use of data sources and capabilities, such as exporting a General Transit Feed Specification (GTFS) feed of a given bus network scenario, that agencies may have access to through existing transit planning tools. Direct demand models are developed to estimate bus ridership at the level of approximately one-mile route-segments and time-of-day periods. This level of analysis provides a more disaggregated evaluation of bus ridership than past direct demand models. The models are sensitive to both route and network improvements. New variables designed to capture the relationship between bus routes, including the competitive and complementary nature of routes, are introduced and incorporated in the model development process. These models are developed for the Washington Metropolitan Area Transit Authority (WMATA). A case study analyzing two scenarios in WMATA's Better Bus Network Redesign (BBNR) is presented, with selected route examples to illustrate how the models capture different types of service changes. These routes fall under three categories: routes with no major service changes, routes with improvements in frequency, and routes with re-routing and other improvements. An open-source multimodal routing algorithm, available through an R package called r5r, is used for travel time analysis. r5r calculates a distribution of door-to-door travel times for a given origin-destination (OD) matrix and returns a selected percentile value from the distribution for each OD pair. The percentile parameter is calibrated through a comparison of estimated travel times and actual travel times recorded in origin-destination-interchange inference (ODX) data. Low percentile values were found to provide travel times close to actual travel times. Additional guidance is provided for interpreting travel times from r5r, and use cases related to calculating travel time impacts between scenarios and evaluating rail competitiveness for a given bus network are explored.

Sensing and Predicting Urban Rail Platform Crowding Using Emerging Data Sources

Thu, 01 May 2025 00:00:00 GMT

Sensing and Predicting Urban Rail Platform Crowding Using Emerging Data Sources Fiorista, Riccardo Rail platform crowding poses serious challenges to passenger safety, operational performance, and service quality in urban rail transit systems. This thesis investigates the short-term forecasting of platform-level crowding, focusing on enhancing prediction accuracy, spatial granularity, and operational interpretability through multi-source data integration. We first employ a gradient-boosted tree regression model (LightGBM) to leverage fare card transaction, vehicle location, weather, and public event data from the Washington Metropolitan Area Transit Authority (WMATA) to forecast platform-level occupancies 15–60 minutes ahead of time. Our results show significant improvements over a WMATA-internal baseline while providing a robust data preparation and prediction pipeline. Subsequently, we explore integrating platform-level CCTV data to overcome the lack of real-time crowding estimates. Using a custom-collected image dataset and three computer vision methods, namely object detection (YOLOv11, RT-DETRv2) and head counting (APGCC), crowd-level classification (Crowd-ViT), and semantic image segmentation (DeepLabV3), we demonstrate that estimated counts from calibrated image segmentation maps enable accurate real-time estimation of platform crowding. Additionally, we show that these estimates can correct and improve 15-minute horizon predictions when incorporated with a stochastic gradient-boosted tree learner such as LightGBMLSS. Finally, we extend the time series modeling framework by incorporating network-wide causal influences through an analysis driven by Empirical Dynamic Modeling and Convergent Cross Mapping. We show that accounting for network effects improves predictive performance, particularly for platforms characterized by regular low-occupancy patterns, improving the prediction of anomalies. The work presented in this thesis extends the existing literature on short-term platform crowding prediction, offering new methodologies to incorporate emerging CCTV data and causal network effects for increased prediction accuracy.

Spatial Thinking as an Analytical Lens for Bilateral International Development: Lessons from the Harbor Reconstruction Project in Jamestown, Accra

Thu, 01 May 2025 00:00:00 GMT

Spatial Thinking as an Analytical Lens for Bilateral International Development: Lessons from the Harbor Reconstruction Project in Jamestown, Accra Avis, Victoria This thesis examines transcalar tensions that emerge from urban infrastructure development projects funded through bilateral foreign assistance mechanisms. Using a mixed-methods case study approach to gather data from a wide variety of historical and contemporary primary and secondary sources, this research centers a harbor revitalization and port reconstruction project in Jamestown, a historic fishing community in Accra, Ghana. Having coordinated plans with the Ghanaian national government, a Chinese state-owned construction firm began working on the port in 2020. In 2024, the revitalized harbor and expanded port were officially handed over to the government of Ghana in a widely attended ceremony. The spatial implications of this physical urban infrastructure project across international, national, municipal, and local levels are complex and interrelated. Therefore, this case study is especially relevant at a historical moment when the nature of bilateral engagement may be undergoing significant transformation. This thesis argues that spatial thinking, a foundational concept in urban planning, is a necessary analytical lens to incorporate within international development practice. Despite its relevance, spatial thinking has not been meaningfully incorporated into international development policy or implementation. Therefore, this thesis seeks to bridge epistemic gaps between urban planning and international development by advancing a spatial thinking framework, adapted for use in international development contexts. In doing so, this thesis envisions a future for bilateral development assistance that delivers equitable and sustainable development outcomes across scales of engagement. This approach, rooted in spatial thinking, intends to respond to local community needs and aspirations, capacitate municipal governments, align with national priorities, and accommodate geopolitical dynamics that facilitate bilateral project implementation.

Order Under Pressure: Structural and Magnetic Characterization at Extreme Stresses

Thu, 01 May 2025 00:00:00 GMT

Order Under Pressure: Structural and Magnetic Characterization at Extreme Stresses Riesel, Eric Alan Mechanical stress is an exquisitely versatile tool for controlling chemical bonding. This multi-dimensional synthetic lever tunes the electronic structure of elements and changes the way that atoms arrange and coordinate to one another. These unique electronic configurations and coordination environments have profound impacts on the properties of materials giving rise to functionality ranging from high-temperature superconductivity to diverse magnetism. Despite over a century of research on solid-state materials over one gigapascal (GPa), experimental and theoretical obstacles remain for structural and physical characterization of complex phases which only persist at these conditions. We begin to address the wide-reaching challenge of structural characterization in complex, bulky sample environments by employing recent advancements in generative artificial intelligence to develop a generalized approach to solving the structure of crystalline solid-state materials. We demonstrate that our model achieves a 42% match rate on a curated set of experimental powder diffraction patterns, and we then use our model to solve the structure of several previously unsolved structures at high pressure. We proceed to focus on a different structural characterization problem: defects which arise exclusively under mechanical stress. We demonstrate that site-disorder is unlikely to occur at room temperature and high pressure in InBi and instead propose a set of defects which explain the X-ray spectra and scattering patterns equally well. Progressing to properties characterization and magnetic ordering at high pressure, we experimentally demonstrate that MnBi2, a compound which does not persist to ambient pressure, is a permanent magnet. Comparing the orbital and spin contributions to the total moment across compounds in the Mn–Bi system, we build up design principles for permanent magnets using heavy main-group elements. The combination of our work in structural and physical characterization at extreme stresses charts a path towards the discovery of functional high-pressure bulk materials and defects.

Coherent Dynamic Nuclear Polarization: Mechanistic Insights and Experimental Advances

Thu, 01 May 2025 00:00:00 GMT

Coherent Dynamic Nuclear Polarization: Mechanistic Insights and Experimental Advances Ouyang, Yifu Dynamic Nuclear Polarization (DNP) enhances the sensitivity of solid-state Nuclear Magnetic Resonance (NMR) by transferring polarization from electrons to nuclei. While traditional continuous-wave (CW) DNP has advanced through improved radical design, the development of pulsed DNP—employing short, high-power microwave bursts—has shown the advantages of coherent spin control. We present both theoretical and experimental investigations aimed at understanding and optimizing polarization transfer. On the theoretical side, we examined multiple DNP mechanisms, including a re-evaluation of the Overhauser effect in insulating solids and a foundational treatment of the chirped solid effect. We also identified a new transfer channel, termed Resonant Mixing, arising from interference effects under off-resonance driving. Building on these insights, we developed a general framework for analyzing amplitude-, phase-, and frequency-modulated pulses. This approach enables the design of hybrid pulse sequences that combine modulation and chirping to produce efficient, selective spin transfer. These sequences maintain high enhancement even at reduced microwave power, thereby improving scalability to high magnetic fields. To test the practical viability of this approach, we designed and evaluated a prototype 400 MHz/263 GHz probe incorporating new resonator and RF technologies. While the initial performance was limited, the system provided a testbed for future high-field pulsed DNP experiments under realistic conditions. Together, these results establish a theoretical and technical foundation for next-generation pulsed DNP, emphasizing coherent spin manipulation, power-efficient design, and applicability to high-field, static-solid NMR systems.

Development of Quantitative Solid-State NMR Methods to Characterize Membrane Proteins

Thu, 01 May 2025 00:00:00 GMT

Development of Quantitative Solid-State NMR Methods to Characterize Membrane Proteins Somberg, Noah H. Membrane proteins are critical components of all cells and viruses. While about one quarter of human proteins are membrane proteins, they constitute half of all drug targets. Despite their importance, membrane proteins are underrepresented among known protein structures, constituting only about 2 percent of the Protein Data Bank. This discrepancy is due to the unique difficulties in studying membrane proteins, which make many techniques commonly used in structural biology extremely challenging. Membrane proteins often display a structural dependence on the local environment. It is therefore essential to have structural biology tools to study these critical proteins in native-like environments. Solid-state Nuclear Magnetic Resonance (NMR) spectroscopy provides one of the few methods available to study the structure and dynamics of membrane proteins directly in the lipid bilayer. Herein, practical and theoretical considerations of dipolar and chemical shift anisotropy recoupling experiments are presented. These experiments were applied to the study of membrane proteins. New experiments and novel analysis techniques were developed, and the results guided biophysical understanding and drug development. Among the membrane proteins of SARS-CoV-2, the Envelope (E) protein is the least understood. E forms a membrane-bound ion channel and is associated with inducing the respiratory symptoms of the disease. The exact oligomeric state of E was not known. The fluorine centerband-only detection of exchange (CODEX) experiment was employed to directly measure the oligomeric state of E in lipid bilayers. The transmembrane domain of E forms a pentamer, while a construct including the ectodomain forms a dimer. Under certain conditions, the pentamers cluster together, forming supramolecular assemblies that may have a unique role in the virus life cycle. New sensitivity-enhanced carbon-fluorine rotational-echo double-resonance (REDOR) experiments are developed and used to investigate the drug binding of E. The small molecule drug hexamethylene amiloride binds to E at the protein-lipid interface. This informed the development of higher affinity inhibitors, which were also shown to bind E at the lipid interface. A novel strategy to identify ligand binding sites of proteins without sequential resonance assignment is presented. The technique uses a computationally efficient second moment approximation to calculate REDOR dephasing, and simulated annealing to explore the associated parameter space. The new methods and advances in quantification and simulation of the REDOR and CODEX experiments enhance the available solid-state NMR toolkit for the study of critical membrane proteins.

Improving the reliability of optical phase change materials-based devices

Thu, 01 May 2025 00:00:00 GMT

Improving the reliability of optical phase change materials-based devices Popescu, Cosmin-Constantin Optical components are part of our daily lives, including vision and camera systems, data transmission in telecommunication, sensing applications, manufacturing and medicine, and more. Compact on-chip integrated optics such as photonic integrated circuits and optical metasurfaces can provide us with the desired functionality, but there is a continuous need for active non-volatile tuning capabilities of these devices. Chalcogenide optical phase change materials (PCM) (e.g. Ge₂Sb₂Te₅) have gathered sustained interest in the past several years in the photonics community exactly due to their potential for non-volatile control of optical signals. Prior work had showcased the integration of PCMs via free-space metal heaters for metasurfaces, demonstrating switching for several tens of cycles. To understand the limiting mechanisms preventing extended cycling of such devices, we have developed a near IR transparent platform on doped silicon-on-insulator for testing both material behavior and device performance, along with the auxiliary code and design needed for such testing. Using this platform, a Ge₂Sb₂Se₄Te-based transmissive metasurface filter was demonstrated with a cycling performance of 1250 cycles. Following, the mechanisms limiting the performance of such devices were explored, providing guidelines to improve their reliability and endurance both at the phase change material scale as well as the accompanying device level. Furthermore, we showcase future potential devices that can be leveraged for PCM photonics, including a theoretical design that avoids free carrier absorption losses from the doped silicon heater by placing the dopants at the node of a resonant mode, limiting their overlap with the regions of high field amplitude, and a matrix array of heaters for higher device functionality. Finally, we point towards areas to focus in order to scale these concepts to commercial applications.

Precision Medicine in Diabetes Using Continuous Glucose Monitoring

Thu, 01 May 2025 00:00:00 GMT

Precision Medicine in Diabetes Using Continuous Glucose Monitoring Healey, Elizabeth Diabetes affects millions of individuals around the world and is a leading cause of death. The risk of serious long-term complications in diabetes can be mitigated through early interventions in the form of medication and behavioral changes. However, the pathophysiology of diabetes and the course of the disease are markedly heterogeneous, making it essential that disease management is tailored to the individual. Continuous glucose monitoring (CGM) helps patients manage their disease through the collection of real-time measurements of interstitial glucose, providing insight into glycemic dynamics that laboratory measurements cannot capture. In this thesis, we investigate how CGM can be used to enable personalized disease management in diabetes using modern methods from machine learning and signal processing. We first investigate a model-based approach to estimate metabolic parameters from CGM data. We show that parameters estimated from daily CGM data correlate with parameters derived from in-clinic laboratory measurements. Then, we explore how the rapidly emerging field of generative artificial intelligence can be integrated into diabetes care through analysis of CGM data. We show how large language model agents hold promising potential to assist patients and clinicians in managing diabetes through the synthesis and narrative summarization of large amounts of CGM data. Finally, we leverage observational CGM data to understand heterogeneity in type 2 diabetes. The work in this thesis shows how modern computational methods in machine learning can enable precision medicine in diabetes by leveraging wearable CGM data for improved disease management.

Structural Insights into Mycobacteriales Galactan Biosynthesis

Thu, 01 May 2025 00:00:00 GMT

Structural Insights into Mycobacteriales Galactan Biosynthesis Carter, Alan Wylde The order Mycobacteriales includes a number of severe human pathogens, including Mycobacterium tuberculosis, the causative agent of tuberculosis and a leading cause of infectious disease-related mortality worldwide. The unique cell wall structure of these bacteria is essential for their viability, and has been studied as a potential target for novel therapeutics development. A key component of the mycobacterial cell wall is the galactan, a 30-40 residue linear polysaccharide of galactofuranose (Galf) with an alternating β(1,5) and β(1,6) linkage pattern, synthesized by the polymerase Galactofuranosyl Transferase 2 (GlfT2). While GlfT2 has been established as a processive polymerase with intrinsic sequence control, the mechanism underlying this activity remains unclear. In the studies presented here, we provide structural insights into Nocardia brasiliensis GlfT2 (NbrGlfT2) using X-ray crystallography and cryo-electron microscopy. We characterize both the acceptor-bound and membrane-embedded structures of NbrGlfT2 and propose three models for its catalysis: Processive Galactan Sliding, Feedback-Regulated Sequence Control, and Membrane Curvature-Mediated Polymerization. Furthermore, we structurally characterize a previously undescribed GlfT2 paralog from Rhodococcus equi, which we term ReqGlfT3. We confirm its galactofuranosyl transferase activity and identify the production of β(1,3) and β(1,5) linkages. These findings offer new insights into GlfT2 and related polymerizing glycosyltransferases, which will provide insights into enzymatic regioselectivity mechanisms and polysaccharide biosynthesis across the bacterial kingdom.

Making Energy Work: Enacting Renewable Transitions in the Deserts of Chile and California

Thu, 01 May 2025 00:00:00 GMT

Making Energy Work: Enacting Renewable Transitions in the Deserts of Chile and California White-Nockleby, Caroline Celeste This dissertation explores how different people enact and engage the “Energy Transition,” a temporal orientation increasingly used to describe a range of activities that operate on and through electricity, renewable energy, and fossil fuels. I ask, given the infrastructural, political, economic, and material-semiotic continuities and imbrications between renewable energies and fossil fuels, how do actors craft, stabilize, and mobilize the [just] renewable energy transition? How, in other words, do people distinguish the activities and projects of the energy transition from continuity and more ordinary change? I also ask, what kind of political work is “transition” – along with its usual modifiers, energy, renewability, and justice – doing in the world? Building on scholarship that explores the history, genealogy, materiality, and political economy of energies and resources, I investigate these questions by analyzing energy transition projects across two region-scale field sites: Antofagasta, Chile and Imperial County, California. I find that self-consciously small-scale technologies like maps, models, and pilot projects are vital to assembling just, renewable resources – and to demarcating particular places and projects as in transition. Though these technologies often aspire to make and move green energy 24/7 and worldwide, they face substantial obstacles to doing so. Their value is, thus, often drawn more from the future they index than their present functionality. I term these temporal indexical technologies “anticipatory devices,” and show that such devices gain significance in relation to the particular forms of expertise that actors draw on to design them. Each chapter, therefore, analyzes a different disciplinary form of expertise in which the concepts of renewability, justice, transition, and energy, aided by various anticipatory devices, take shape: cartography (Chapters 1 and 2), chemistry (Chapter 3), engineering (Chapter 4), and economics (Chapter 5). Ultimately, I find that energy transition is often treated as a universalizing, singular narrative, which can shape and limit the scope of climate mitigation projects. Profit motives often incentivize corporate actors to design projects to align with the more temporary kinds of transitions that have long been constitutive of capitalism, even though it is longer-term changes that will most effectively mitigate climate change. The same is true for renewability, which can easily be articulated as an ideal that supports visions for unfettered capitalist growth. Moreover, approaches that treat “transition” as universal can also easily echo and reinvigorate an evolutionist approach to time, in which places and countries compare their relative advancement towards carbon neutrality along a single, teleological temporal axis. Yet I also encountered many actors engaged in more situated projects that attended to local histories of land use, industry, and power. These projects pluralized transition – or did not use the term at all – offering situated and distributive visions of energetic change that might enable more regenerative futures to germinate.

Solid-State NMR Characterization of a PET Ligand Binding Sites in AD Tau Fibrils

Thu, 01 May 2025 00:00:00 GMT

Solid-State NMR Characterization of a PET Ligand Binding Sites in AD Tau Fibrils Angehrn Rodas, Frida Nicole Aggregation of the tau protein into fibrils is a key feature of Alzheimer's disease (AD) and many other neurodegenerative disorders. Developing small molecules that bind these tau fibrils is important for the diagnosis and treatment of tauopathies. This thesis revolves around a study on the binding sites of a positron emission tomography (PET) ligand, PI-2620, to a recombinant tau construct that adopts the C-shaped AD fold. Using solid state NMR experiments in combination with other techniques such as Transmission Electron microscopy (TEM) as well as docking simulations allowed a better understanding of the binding sites of this PET agent. Specifically, 13C-19F REDOR experiments were used to identify nearby residues to the ligand. PI-2620 was found to bind two primary sites within the C-shaped structure. The docking simulations allowed the proposition of several possible binding poses. Additional 2D NMR experiments suggest that PI-2620 alters the protofilament interfaces. The stoichiometry of PI-2620 binding to tau fibrils was determined to be approximately 20 mol%, with varying degrees of ligand mobility. These findings offer insights into the interaction of this PET tracer with tau fibrils and have implications for the design of improved imaging agents.

Demand‑Driven Decarbonization: Impact of Voluntary 24/7 Low-Carbon Power Procurement

Thu, 01 May 2025 00:00:00 GMT

Demand‑Driven Decarbonization: Impact of Voluntary 24/7 Low-Carbon Power Procurement Ali, Adam This thesis examines the impact of voluntary 24/7 (hourly) low-carbon power procurement on grid-wide emissions and investment strategies in generation technologies. Recognizing the growing number of businesses and government agencies making voluntary commitments to reduce greenhouse gas emissions (GHGs) through increased procurement of low-carbon power, this study investigates the effectiveness of these commitments, particularly those aiming for hourly matching of low-carbon energy with consumption. This study employs GenX, an open-source capacity expansion model, to simulate an electricity market with two classes of buyers. Buyers in one class commit to reduce the carbon intensity of their electricity procurement by some amount, while buyers in the other class procure electricity at minimum cost without any regard to carbon emissions. This setup allows for a detailed examination of how different levels of ambition in voluntary hourly low-carbon commitments influence the electricity system and investment strategies. The study tests both a simpler model without storage and demand-response capabilities and a more complex model that incorporates these elements to assess their impact on meeting hourly clean energy targets. Our findings suggest that at low to moderate ambition levels of hourly low-carbon electricity procurement, the buyers with voluntary commitments can primarily "reshuffle" built low-carbon generation without incentivizing new clean capacity additions or achieving measurable reductions in system-wide emissions. Significant shifts in generation investments and decreases in total carbon emissions are observed only when commitments exceed a critical threshold, ranging from approximately 70% to 96%, depending on the facts of the system, which happen to be reflected in different model set-ups. Even then, cost-minimizing behavior in voluntary procurement can distort investment, spurring excessive wind and solar builds that exceed what a least‑cost, socially-optimal zero‑carbon portfolio would require. In conclusion, for voluntary 24/7 procurement to cut emissions materially—and avoid misallocating capital—either ambition must be extremely high or participation must broaden enough to share costs and benefits. Otherwise, committed buyers bear steep costs, non‑participants enjoy spill‑over gains, and the system drifts toward a sub‑optimal technology mix.

Clarifying Decision Making Processes: Tools for Interdependency Modeling

Thu, 01 May 2025 00:00:00 GMT

Clarifying Decision Making Processes: Tools for Interdependency Modeling Baker, Ellie F. Tools for problem specification in AI Decision making are underdeveloped at present. I propose two new tools for this purpose; first, a model of AI Decision Making, which supports problem identification and mitigation. Second, a Bill of Assumptions for Data Production. Data is an important component of AI Decision Making Systems, and data is necessarily produced by making a series of assumptions. My Bill of Assumptions for Data Production is a new approach to communicating these assumptions that facilitates collaboration, data transparency, and reduction of harmful bias. I illustrate this new approach by developing a dataset that estimates the distribution of Government education spending in the US across income deciles. My dataset informs existing Distributional National Accounts (DINA), which are a primary measure of income inequality in the US (Piketty et al., 2018). My estimate shows Government education spending is more progressive than assumed in current DINA. Furthermore, I show that removing federal education funding to postsecondary institutions would produce substantial harm.

Containerless measurement and thermodynamic prediction of the physical properties of liquid steels

Thu, 01 May 2025 00:00:00 GMT

Containerless measurement and thermodynamic prediction of the physical properties of liquid steels Benderly-Kremen, Ethan Incomplete control over macrosegregation during steel solidification hinders the development of novel steel alloys and applications by limiting compositions which can be successfully cast. Analysis of macrosegregation at the solidification front is aided by study of the liquid state via fluid mechanics, which can place bounds on when macrosegregation can occur. This non-dimensional analysis requires knowledge of the physical properties of the liquid steel: density, surface tension, viscosity- and how they change with composition as solutes are rejected from solid at the solidification front. Macrosegregation is most pronounced in ferrous liquids containing light, non-metallic species, i.e. carbon, oxygen, and sulfur. Yet, existing literature models for predicting the physical properties of liquid alloys are incapable of accounting for these interstitial species inside an iron lattice. Additionally, direct experimental measurement of these properties is hindered by the requisite high temperature, high reactivity of the melt, and the vast composition space of steel alloys. Herein, both the experimental and modelling challenges are introduced and addressed. An experimental technique using a floating zone furnace, pendant drop geometry, high-speed camera, and video segmentation was developed for simultaneous, containerless, high-throughput measurement of the physical properties of liquid steel samples. The central atom model, a multicomponent solution model, is extended to investigate the statistical structure of alloys consistent with their energetics and solution thermodynamics. This allows liquid structure determination from thermochemical measurements, bypassing structural and atomistic modeling challenges of high-temperature liquid systems. These methods and models are explored on the binary systems of iron-nickel, the major substitutional alloying element in steel, and iron-carbon, the major interstitial species. Results demonstrate successful liquid property measurement at experimental rates far exceeding traditional high-temperature research and introduce a basis for a unified treatment of thermodynamic and physical properties in multicomponent alloy melts.

Transforming geospatial textual data into narrative storytelling visualization

Thu, 01 May 2025 00:00:00 GMT

Transforming geospatial textual data into narrative storytelling visualization Ma, Ruixian Current large language models (LLMs) often struggle to integrate geospatial data into dynamic, interactive visualizations, relying instead on text-based outputs. This limitation hinders the full potential of geospatial data to convey complex information through narrativedriven communication, making it difficult for users to interpret the data easily. Meanwhile, existing data visualization tools typically depend on static dashboards and rigid scientific formats, which have a steep learning curve and lack engagement through narrative elements. Audiences, however, are increasingly drawn to story-driven presentations, as seen in platforms pioneered by the MIT Senseable City Lab, and widely popularized by The New York Times and the Washington Post, which use narrative data visualization formats to attract and immerse readers. This gap between the capabilities of current LLM-based tools and users’ preferences presents a unique opportunity to develop a narrative-based geospatial visualization tool that meets these needs. This tool could transform how we communicate spatial data, particularly in fields such as journalism, travel planning, and urban planning, where the ability to convey complex patterns in an engaging manner is essential.

Quantifying the Post-Pandemic Urban Activity & Mobility Regime: Implications for Adaptation and Future Planning of Cities and Public Transit Systems

Thu, 01 May 2025 00:00:00 GMT

Quantifying the Post-Pandemic Urban Activity & Mobility Regime: Implications for Adaptation and Future Planning of Cities and Public Transit Systems Leong, Chee Weng Michael Between 2019 and 2022, a pattern break during the COVID-19 pandemic introduced consequential changes to the trajectory of urban activity and mobility patterns. This thesis advances both theoretical and practical understandings of this evolving post-pandmemic regime of activity and mobility, as well as its implications for the future of cities and public transit systems, using high-resolution location-based services data and a case study within the Washington, DC metropolitan area. First, a custom analysis framework is developed where geographical units - subcenters and neighborhoods - are designed to provide insight at an interpretable scale that corresponds to policy and business decision making. Second, a custom suite of twelve mobility metrics are curated to distill the applicability of post-pandemic changes in travel patterns to business problems (site selection, network planning, and operations planning) and societal outcomes (social fabric, quality of life, and environmental sustainability). To complement spatial analysis, these metrics are also regressed on socio-economic attributes to provide greater explanatory power. Lastly, key trends in post-pandemic activity and mobility are distilled into eight mega-trends, and their implications for the adaptation of public transportation systems and future urban development are discussed, including complexity from divergent definitions of success among different stakeholders.

Small Stores, Big Obstacles: Understanding Constraints and Opportunities for Micro-retail Firms

Thu, 01 May 2025 00:00:00 GMT

Small Stores, Big Obstacles: Understanding Constraints and Opportunities for Micro-retail Firms Cervantes Gil, Sergio Yael Micro and small enterprises (MSEs), particularly informal micro-retailers known as nanostores, play a vital role in developing economies but remain largely underserved by traditional financial institutions and overlooked in economic policy. In Mexico, nanostores account for more than 95% of businesses and over 10% of national employment, yet face high closure rates, low productivity, and limited access to formal credit. This thesis asks: What structural and contextual factors determine the survival and performance of nanostores, and how can policy better support high-potential firms within this segment? To answer this, the study constructs a longitudinal panel of nanostores using microdata from the Mexican Economic Census (2009, 2014, and 2019), and combines it with municipality-level contextual data including crime, infrastructure, unemployment, electricity costs, and business regulations. It applies survival models to estimate firm closure dynamics and implements a misallocation framework to quantify distortions in capital and labor usage. The results reveal that misallocation—particularly of capital—is pervasive and systematically linked to institutional weaknesses and credit access constraints. In response to the limited real-time data available for this sector, the thesis proposes the LIFT Performance Index, developed by the MIT Low-Income Firms Transformation Lab (MIT LIFT Lab), as a diffusion-based tool for monitoring micro-retailers’ business sentiments using structured operational surveys. A pilot implementation in Argentina demonstrates the index’s potential to generate timely and actionable insights for policymakers and private stakeholders. Overall, this work contributes a novel empirical foundation for understanding heterogeneity within the micro-retail sector and offers a scalable framework for designing targeted, data-driven interventions to support inclusive economic development.

Chemical Probes and Strategies to Study Mycobacterial Cell Envelope Assembly

Thu, 01 May 2025 00:00:00 GMT

Chemical Probes and Strategies to Study Mycobacterial Cell Envelope Assembly Lee, So Young The cell envelope of Mycobacterium tuberculosis (Mtb) is central to its pathogenicity, immune evasion, and intrinsic drug resistance. While the importance of its glycan components is well recognized, their structural intricacies have hindered efforts to directly perturb and investigate their function. In this work, we discuss chemical approaches to study and manipulate mycobacterial cell envelope biosynthesis. Specifically, we present biosynthetic glycan labeling strategies that leverage the activity of native glycosyltransferases to probe arabinogalactan and mannose-containing glycolipids. Building upon prior work using lipid-linked probes to label mycobacterial arabinan, Chapter 2 details the development of azido-functionalized farnesyl phosphoryl mannose (AzFPM) probes that mimic native polyprenyl-phosphoryl donors and selectively label mannose-containing glycolipids in live mycobacteria. Chapter 3 showcases how these probes enable glycan substructure-specific labeling and biochemical enrichment of glycolipids across Corynebacterium glutamicum, Mycobacterium smegmatis, and Mtb. This strategy provides a platform to study glycolipid dynamics in wild-type cells, a task previously hindered by the lack of selective labeling tools. In Chapter 4, we further interrogate endogenous glycan biosynthesis by applying biosynthetic labeling probes in C. glutamicum. Perturbation of arabinan structure by probe incorporation led to impaired cell wall integrity and growth defects. In glycosyltransferase deletion strains, altered probe incorporation patterns revealed enzyme-specific roles in glycan assembly and architecture. Beyond novel labeling strategies, Chapter 5 describes the development of targeted inhibitors of galactan biosynthesis, an essential yet underexplored component of the mycobacterial cell wall. We employed a prodrug strategy to inhibit UDP-galactopyranose mutase (UGM), which catalyzes the committed step in Galf production. To overcome delivery challenges, we designed amide prodrugs activated intracellularly by amidases. One prodrug exhibited improved efficacy in Mtb, providing a promising lead for antibiotic development. Collectively, these studies establish biosynthetic labeling and targeted galactan inhibition as powerful tools for dissecting the structure and function of the mycobacterial cell envelope, offering new avenues for developing chemical probes and therapeutics against tuberculosis.

Photonic design for chemical analysis

Thu, 01 May 2025 00:00:00 GMT

Photonic design for chemical analysis Ma, Wenchao With light being manipulated at subwavelength scales, photonic design has been explored for various applications. In this dissertation, we investigate potential application of photonic design to chemical analysis, with a focus on spectrometry and chemical sensing. First, we demonstrate inverse design of single-layer metasurfaces with shape optimization. Each of the designed metasurfaces simultaneously focuses light and shapes the spectra of focused light without using any filters. Thus, both spatial and spectral properties of the meta-optics are engineered. We chose the color matching functions of the CIE 1931 XYZ color space as the target spectral shapes and a distant region with a finite size as the focal area. We then present an inverse-design approach for computational spectrometers, in which the scattering media are topology-optimized to achieve higher robustness in inference, without the need of a training set of spectra and noise. Our approach also allows the selection of the inference algorithm to be decoupled from that of the scatterer. For smooth spectra, we devise a regularized reconstruction algorithm based on Chebyshev interpolation, which yields higher accuracy compared with conventional treatment in which the spectra are sampled at equally spaced frequencies or wavelengths with equal weights. Our approaches are numerically demonstrated via inverse design of integrated computational spectrometers and reconstruction of example spectra. The inverse-designed spectrometer exhibits significantly better performance in the presence of noise than its counterparts with random scatterers. Furthermore, we discuss chemical detection using optical resonances, which can increase the sensitivity of measurements to material perturbations and also accelerate photochemical reactions. We show that these two effects can be combined multiplicatively, to enhance the detection via weak/low-concentration photochemical reactions far beyond what could previously be attained. For an optical resonance with a quality factor Q, the sensitivity of our detection scheme is enhanced by ~ Q² (where ~ denotes approximate proportionality), as demonstrated by both theoretical arguments and numerical simulations of a simple optical grating resonance coupled with reaction-diffusion equations. Such an approach opens a door to further improvements by careful design of the resonance: even a 3-parameter optimization of the grating resonance yields an additional ≈ 7 × improvement. Finally, regarding linear electromagnetic systems possessing time-reversal symmetry, we present an approach to bound ratios of internal fields excited from different ports, using only the scattering matrix, improving upon previous related bounds by Sounas and Alù (2017). By reciprocity, emitted-wave amplitudes from internal dipole sources are bounded in a similar way. When applied to coupled-resonant systems, our method constrains ratios of resonant coupling/decay coefficients. We also obtain a relation for the relative phase of fields excited from the two ports and the ratio of field intensities in a two-port system. In addition, although lossy systems do not have time-reversal symmetry, we can still approximately bound loss-induced non-unitarity of the S matrix using only the lossless S matrix. We show numerical validations of the near-tightness of our bounds in various scattering systems.

Sulfidation of Ternary Oxides: A Thermodynamic and Experimental Study Toward Selective Metal Extraction

Thu, 01 May 2025 00:00:00 GMT

Sulfidation of Ternary Oxides: A Thermodynamic and Experimental Study Toward Selective Metal Extraction Boury, Charles A. Conventional metal refining techniques face growing challenges due to increasing ore complexity and their limited ability to accommodate post-consumer recycling feedstocks. Sulfidation is a promising pyrometallurgical approach for the selective separation and recovery of critical elements from such complex feedstocks. This thesis presents a chemical thermodynamic framework for the sulfidation of divalent alkaline earth and transition metal ternary oxides of titanates, molybdates, tungstates, niobates, and tantalates. Modified predominance diagrams were constructed to determine the possible sulfidation outcomes, and a sensitivity analysis on the input thermodynamic data was performed to assess their impact on the outcome of sulfidation. A high-temperature apparatus was designed and tested to compare predicted and observed sulfidation behavior. Together, the model and experimental apparatus provide a new experimental method to estimate the high-temperature Gibbs energy of multicomponent oxides. Application to current chemical metallurgy challenges, including the recycling of tantalum-based capacitors and the refining of tungsten-bearing ores, underscore sulfidation as a powerful step to support new processing route for sustainable metal recovery.

Maximizing Flexibility and Efficacy of Undersea Wireless Power Transfer Systems

Thu, 01 May 2025 00:00:00 GMT

Maximizing Flexibility and Efficacy of Undersea Wireless Power Transfer Systems Gess, Derek Autonomous underwater vehicles (AUVs) are an ever-increasingly essential tool for ocean-based applications, whether it be scientifically, economically, or militarily. To advance the capabilities of AUVs, it is crucial to improve the mission time and length of these vehicles. One proposed way to achieve this is with remote undersea wireless power transfer (WPT) systems to allow AUV charging from remote areas of the ocean floor. While there has been significant research in WPT system design, these projects often tailor the design specifications towards a specific AUV shape, size, or power requirement. These point designs have wildly different power outputs, efficiencies, coupling coefficients, sizes, and more, making it difficult to understand how the design parameters affect each of these properties. This paper aims to address this knowledge gap in current undersea WPT systems by designing an equivalent circuit framework for a WPT system with a targeted power output of ~1 kW to show how design parameters such as input voltage, coil size, transfer gap, coupling coefficient, and load resistance affect the power output and efficiency of the charger. Furthermore, the effects of misalignment in vertical and lateral directions for two separate compensation networks – series-series (SS) and series-parallel (SP) – are compared to determine which compensation network would perform best under specified circumstances. The paper then addresses the losses associated with a conductive environment by coupling the circuit model with an electric field model in seawater. The impact of undersea losses on system metrics is quantified, showing a 3% decrease in efficiency as compared to in air. Finally, the study investigates the use of magnetic cores in WPT systems for EM shielding and field-shaping characteristics. A design methodology is introduced to rank material properties based on the desired system performance characteristics. Suggested materials are then chosen according to this ranking and tested using the models derived in the study. By mapping both electrical and magnetic-core design spaces in a conductive seawater environment, this thesis delivers a unified methodology for designing scalable, efficient undersea wireless chargers.

Fundamentals, Voltage Control and Novel Application of Exchange Bias in Magnetic Thin Films

Thu, 01 May 2025 00:00:00 GMT

Fundamentals, Voltage Control and Novel Application of Exchange Bias in Magnetic Thin Films Hasan, Muhammad Usama The past half-century has seen remarkable advances in microelectronics, but as transistors approach their physical limits, there is a growing need for beyond-CMOS technologies. Spintronics, which aims to utilize the electron's spin in magnetic thin films for data storage and manipulation, is a promising alternative. Among the rich physical interactions that appear in magnetic thin films, the exchange-bias (EB) effect is essential for many spintronic devices. EB is an effect that arises at ferromagnet/antiferromagnet interfaces, which imposes an internal field on the ferromagnet, enhancing the range of functionalities that can be derived from devices. This thesis explores EB in Co/Co₁₋ₓNiₓO systems at multiple levels, from fundamental understanding to its manipulation and applications. First, we introduce a new model to predict EB in polycrystalline antiferromagnetic thin films and validate it with experimental data. Second, we tackle another fundamental aspect – disentangling the effects of EB on nucleation and propagation of magnetization reversal. We discover that nucleation and propagation EB can be unequal and demonstrate how that can lead to unexpected behavior of the system, including having asymmetric hysteresis loops. Building on these insights, we demonstrate voltage-controlled ionic gating to manipulate EB, achieving cyclic toggling of the EB sign in a ferrimagnetic system, where the magnetization direction is fully determined by the gating state. Furthermore, by targeting the antiferromagnet directly, we discover EB enhancement up to 100%, which can be explained with the help of the model developed earlier. We demonstrate sub-millisecond and analog operation in this system. Finally, a new approach to improving bit-stability whilst preserving performance in magnetic racetrack memory is proposed which involves incorporating an EB layer with the right properties into the track. The benefits obtained from this strategy can help push this next-gen memory device closer to commercialization. We believe the findings in this thesis substantially extend the state-of-the-art in terms of basic understanding of EB, ways of EB manipulation and unexplored use-cases of EB, paving the way for new functionalities in spintronic devices applicable for non-conventional computing paradigms or next-gen memory devices.

Essays on Urban Resilience to Environmental and Health Risks

Thu, 01 May 2025 00:00:00 GMT

Essays on Urban Resilience to Environmental and Health Risks Fan, Yichun Cities today face growing environmental and health threats due to climate change. Building urban resilience requires understanding the complex interplays between environmental and social systems that account for adaptation dynamics. Using new data, computational tools, and economic analysis, this thesis explores how people and places adapt to environmental risks and the implications for urban policy and infrastructure planning. Chapter 1 examines how the financing structure of climate resilience infrastructure impacts long-term economic dynamics. Using satellite imagery to develop new performance metrics for U.S. flood protection levees, I find that decentralized financing of infrastructure maintenance creates a feedback loop: lower housing values and property tax revenues reduce fiscal capacity for levee maintenance, which increases levee failure risk and further depresses housing values. These feedback dynamics reinforce under-maintenance and perpetuate spatial inequality. Chapter 2 analyzes the social cost of behavioral adaptation. Leveraging 27 million fitness app exercise records and quasi-experimental designs, I find that heavy air pollution reduces outdoor exercise likelihood by 28%, with information and risk awareness as key moderators. This behavioral response results in significant health costs often overlooked in environmental health studies. Chapter 3 explores the role of subjective traits in predicting adaptation behavior. Applying Natural Language Processing to social media posts from 500,000 users, we classify individual fear types and find that pre-pandemic fearfulness strongly predicts social distancing behavior during COVID-19. This project provides a scalable tool for measuring unobserved subjective traits to predict behaviors under risk and target interventions.

City as Seed: The Urban Resonance Field and the Case for Sonic Awareness in Ecological Renewal

Thu, 01 May 2025 00:00:00 GMT

City as Seed: The Urban Resonance Field and the Case for Sonic Awareness in Ecological Renewal Navarro, Cadine Seeds, the “abominable mystery” (Darwin, 1897), hold our past and potential future. They also hold sound. Much like cities, they are sites of growth, transformation, and resilience. This thesis draws parallels between laboratory research on the sensing capacities of seeds and embodied experiences of sensing within urban landscapes, exploring how living systems interact with sound and vibration. Through both scientific and poetic approaches, it examines how seeds respond to sonic environments and how this sensitivity can inform human engagement with acoustics in the urban context. The investigation of intangible forces, vibration, resonance, and sound reveals a shared responsiveness between seeds and cities, documented through graphs, sound spectra, and reflective narratives that bridge science and art. Focusing on sound as a strategic lens, this work brings attention to often-overlooked sensory domains, inspiring a more ecologically and socially responsive urbanism. Ultimately, it advocates for practices of deeper listening as a method to engage openly and imaginatively with human and nonhuman worlds, and to reimagine urban environments as spaces of attunement, dialogue, and co-existence.

Monitoring and Treating Neurological Conditions Through Focal Interfacing with the Brain

Thu, 01 May 2025 00:00:00 GMT

Monitoring and Treating Neurological Conditions Through Focal Interfacing with the Brain Jackson, Hannah Dale Neurological dysregulation serves as the fundamental basis for a spectrum of debilitating disorders such as Parkinson's disease and epilepsy. Despite considerable efforts, our current comprehension of these disorders and ability to treat them remains limited. Neurochemical sampling of the affected tissue can be used to monitor pathological states, but existing tools are limited by tissue reactivity and suboptimal spatiotemporal resolution. Additionally, methods for treating neurological disorders predominantly rely on systemic drug administration which is hampered by inadequate targeting and off-target effects. There is a need for minimally invasive modalities to both monitor and treat neurological disorders that have high spatial resolution, maintain chronic functionality, and preserve overall brain function. This thesis presents the development and implementation of neural implants capable of both infusing and sampling sub-microliter volumes of fluid with exceptional spatial precision. These implants utilize micron-scale technology to minimize scarring following implantation and allow for sustained chronic functionality. We use these devices to answer two key questions: (1) Can the localized delivery of drugs to specific neural circuits provide effective treatment for neurological diseases? and (2) Can micro-invasive sampling of brain interstitial fluid facilitate disease diagnosis and monitoring? We assessed our ability to treat focal epilepsy with this platform by delivering antiseizure medications directly to the seizure focus in a mouse model of temporal lobe epilepsy. We found that localized drug delivery effectively suppressed seizure activity without adverse effects. We also explored micro-invasive, membrane-free sampling of interstitial fluid from different brain regions using our device. We detected hundreds of distinct proteins from minute sample volumes with high spatial resolution and minimal tissue damage. The results from both studies highlight the platform’s potential for targeted drug delivery and biomarker detection across a variety of disease states.

Toward sequence-to-structure predictions of chromatin: Generative AI sheds light on genome organization

Thu, 01 May 2025 00:00:00 GMT

Toward sequence-to-structure predictions of chromatin: Generative AI sheds light on genome organization Schuette, Greg The secrets of the genome have captivated scientists for well over a century, though the active role its spatial organization plays in gene regulation, cell determination, and disease formation has become clear only in recent decades. Significant strides have been made toward characterizing and understanding three-dimensional genome organization, but the scale, complexity, and heterogeneity of the genome and nuclear environment complicate investigations into this system. This thesis alleviates these challenges and holds the potential to accelerate genome organization research by presenting several methodological advances. An efficient Hi-C inversion algorithm appears first. This technique extracts pairwise contact potentials from experimental Hi-C data, uncovering mechanistic details obscured by the correlation between Hi-C contact probabilities. This required the development of a spin-glass model of chromatin and the derivation of a corresponding model inversion; the model may find use in further theoretical studies of chromatin, while the inversion can be applied more broadly. The inversion successfully revealed the location of chromatin loop anchors, supported the phase separation formation of chromatin compartments, and parameterized polymer models that reproduced the experimental Hi-C data with reasonable accuracy. The focus then shifts toward ChromoGen, a generative AI model that predicts three-dimensional chromatin structures directly from DNA sequence and chromatin accessibility data. ChromoGen provided biologically accurate structural ensembles throughout the genome of two cell types, including one omitted from its training data. This transferability suggests that ChromoGen can provide access to the organization of chromatin in a wide variety of cell types while only relying on widely available sequencing data. Afterward, we discuss several strategies to extend ChromoGen to full-chromosome structure prediction tasks. Preliminary results suggest that the technology of today can provide this capability, as we have generated physical chromosome conformations for mouse chromosomes, although sequencing data did not guide this generative process. Correspondingly, we explore the possibility of incorporating a multimodal model with ChromoGen, allowing it to condition structure generation on a wide variety of data types. Success in this area could enable true de novo structure prediction, greatly simplifying research aiming to understand the relationship between sequence, structure, and cellular function while also accelerating the development of treatments for diseases that implicate chromatin dysregulation.

Reparative Urban Science: Challenging the Myth of Neutrality and Crafting Data-Driven Narratives

Thu, 01 May 2025 00:00:00 GMT

Reparative Urban Science: Challenging the Myth of Neutrality and Crafting Data-Driven Narratives So, Wonyoung This dissertation constructs a distinctive lens on how we should see urban technology in the context of a long history of systemic racism, and how we can take a reparative approach to intervene in contemporary situations of racial inequality with technology/data as a method to address systemic racism. The current discourse of urban science often puts an emphasis on newly available (and big) data, primarily values methodologies of “hard” sciences such as physics, computer science, and mathematics, and evolves to incorporate the latest technologies and analytic methods including machine learning and artificial intelligence. However, the role of urban science and analytics that “move[s] beyond analysis” has not been extensively theorized. In particular, the relationship between urban technologies, white supremacy, and racial capitalism has not been extensively studied. Nonetheless, the impact of the applications of such “urban analysis” on people’s lives has been substantial. Building on planning scholars’ calls for reparative planning and emerging discourses of “algorithmic reparation,” this dissertation proposes a normative framework of reparative urban science that challenges whiteness in urban science and embraces the epistemologies and methodologies of reparations. The dissertation follows a three-paper structure, with the first paper serving as the theoretical framework for two empirical studies, and includes a concluding chapter. The first paper introduces the overarching theory of reparative urban science, identifying three mechanisms—formalizing, context removal and legitimization, and penalization—through which urban technologies perpetuate historical inequalities under a race-neutral guise. It then proposes reparative methodologies, including algorithmic auditing, crowd-sourced community data collection, and algorithms designed to simulate and deliberate reparative futures. The second and third papers demonstrate reparative urban science in action, exemplifying these methodologies. The second paper investigates tenant screening services and landlord decision-making. It reveals the mechanisms how tenant screening algorithms contribute to obscuring historical racial biases, and how landlords interact with them to exert harms. The third paper evaluates the reparative potential of housing programs using algorithmic methods, particularly comparing race-neutral versus race-conscious Special Purpose Credit Programs (SPCPs). It demonstrates that race-conscious SPCPs could significantly reduce the racial housing wealth gap than race-neutral ones, showing race-conscious policies as potential reparative tools. The concluding chapter explores theoretical and practical considerations of housing reparations through the lens of reparative justice, arguing for a deeper acknowledgment of the historical and structural harms related to land and property. Overall, this dissertation seeks to reorient urban science toward justice and repair, envisioning a transformative path forward that actively confronts historical harms and promotes healing and equity in urban futures.

Growth-Induced Cation Order and Magnetic Anisotropy Engineering in Iron Garnet Thin Films

Thu, 01 May 2025 00:00:00 GMT

Growth-Induced Cation Order and Magnetic Anisotropy Engineering in Iron Garnet Thin Films Kaczmarek, Allison C. At its heart, Materials Science and Engineering is a discipline seeking to advance technologies by improving the materials that make them. Currently, the development of electronic devices is limited by the need for materials that enhance speed, reduce size, and improve energy efficiency. Spin-based memory devices, which encode data in a material’s magnetic state, offer a promising solution. Magnetic memory technologies are widespread today, powering devices such as memory disks, tapes, and magnetic random access memory (MRAM). While garnet materials have long been studied for these applications, they have faced challenges in becoming adoptable technologies. However, with advanced research techniques and a deeper understanding of material behaviors, magnetic garnets are experiencing a renaissance. This thesis explores the engineering of iron garnet thin films for next-generation spin-based memory applications. The work presented advances the understanding of non-equilibrium growth, characterization, and engineering of iron garnet thin films and their magnetic properties, emphasizing kinetic phenomena that govern atomic organization beyond classic ordering of unit cells and emergent magnetic anisotropy. In a composition series of europium-thulium iron garnet (EuTmIG) films, experiment confirms the 50-year-old theory of cation site preference of Eu and Tm, demonstrating that enhanced magnetic anisotropy, named ’magnetotaxial anisotropy’, is linked to cation ordering during growth. These findings lay the foundation for anisotropy engineering by cation order. Further studies investigate the effects of film formulation, growth kinetics, and post-growth annealing on structural ordering and magnetotaxial anisotropy. In bismuth-yttrium iron garnet (BiYIG) films, a linear relationship between Bi-Y ordering, magnetic anisotropy, and substrate-lattice mismatch provides deeper insight into the forces that drive cation ordering. Annealing is shown to further enhance magnetic anisotropy in these films. In lutetium-yttrium iron garnet (LuYIG) films, the laser pulse rate during growth by pulsed laser deposition is shown to influence Lu-Y ordering and magnetic anisotropy, reinforcing the kinetic nature of the cation ordering. The findings of this thesis contribute to the fundamental understanding of cation ordering in complex oxide films and provide a framework for engineering and characterizing garnet materials, enabling the future development of new spintronic devices.

Controlling and probing nonlinear collective mode dynamics in quantum materials

Thu, 01 May 2025 00:00:00 GMT

Controlling and probing nonlinear collective mode dynamics in quantum materials Zhang, Zhuquan Tailored laser pulses offer a powerful means of driving materials out of equilibrium by selectively addressing specific degrees of freedom. In particular, the excitation of low-energy collective modes in solids—such as lattice vibrations (phonons) and spin precessions (magnons)—to large amplitudes opens fundamentally new pathways for controlling and probing material properties that are otherwise inaccessible under thermal equilibrium conditions. In this regime, both the nonlinear interactions between light and matter and the intrinsic nonlinear dynamics of the driven modes present significant challenges for understanding the underlying mechanisms and for realizing potential applications. This dissertation centers on two major themes: (1) probing equilibrium properties of materials via nonlinear light-matter interactions; and (2) unveiling emergent phenomena hidden in equilibrium by driving collective modes far from equilibrium. I begin by providing an overview of recent advances in controlling and probing quantum materials out of equilibrium, followed by a discussion of the theoretical frameworks and experimental methodologies used to interrogate collective excitations. Building on this foundation, I present two studies demonstrating how terahertz Raman excitation can reveal distinct spectroscopic signatures of material states. Subsequently, I focus on coherent nonlinear magnon-magnon interactions in canted antiferromagnets, induced by tailored terahertz fields. In these experiments, we demonstrate a unidirectional magnon upconversion process and identify correlated magnonic responses at both the sum and difference frequencies of the interacting modes. We achieve parametric amplification of magnon coherence by tuning the magnonic difference-frequency generation into resonance with a low-frequency magnon. Furthermore, by increasing the driving field strength to access a far-from-equilibrium regime, we uncover spectroscopic signatures of non-perturbative dynamics marked by strong magnon self-interactions. Finally, I present an example in which spatially heterogeneous responses of electromagnon modes in a van der Waals multiferroic are revealed through terahertz photon echo measurements. Together, these results highlight how tailored light-matter interactions can be leveraged to probe, control, and manipulate material degrees of freedom, both in and out of equilibrium.

Decoding Brain Somatic Mosaicism with New Single-Cell Copy Number Analysis Methods

Thu, 01 May 2025 00:00:00 GMT

Decoding Brain Somatic Mosaicism with New Single-Cell Copy Number Analysis Methods Zhao, Yifan Copy number variants (CNVs) represent a significant but understudied form of somatic variation in the human brain, with potential implications for neurodevelopment, aging and disease. While single-cell whole-genome sequencing (scWGS) enables genome-wide profiling at single-cell resolution, existing computational methods struggle to accurately detect non-clonal CNVs, limiting our understanding of genomic mosaicism in the brain. In this thesis, I present two novel and complementary computational approaches for high-resolution CNV analysis in single cells. The first, HiScanner, is a CNV detection method that integrates single-cell assay-specific characteristics and introduces innovations in bin size optimization, read depth normalization, and joint segmentation across cells. Through extensive benchmarking experiments, I demonstrate HiScanner’s superior performance compared to existing tools. The second is a validation method that leverages unique molecular patterns from tagmentation-based scWGS, representing the first tool that exploits fragment overlap patterns to corroborate CNV predictions. I then apply these tools to investigate CNVs in three biological contexts: tumor evolution in paired initial and recurrent meningiomas, age-related genomic changes in neurotypical human brains, and developmental patterns in fetal and postnatal brain tissues. By analyzing both scWGS and multimodal single-cell data (paired RNA-seq and ATAC-seq), I characterize cell-type-specific CNV patterns and their potential functional implications. This work establishes a robust framework for studying somatic CNVs at single-cell resolution and provides insights into genomic instability in brain development, aging, and disease.

Critical Vulnerabilities of AI in Latin America

Thu, 01 May 2025 00:00:00 GMT

Critical Vulnerabilities of AI in Latin America Dobles Camargo, Claudia Artificial Intelligence (AI) is rapidly reshaping societies, economies, and governance systems worldwide. While it offers tools for addressing critical challenges—such as climate change, health care, and educational inequity—it also risks deepening historical inequalities, undermining democratic institutions, and exacerbating global technological dependencies if not ethically governed. Latin America faces unique vulnerabilities for the development and use of AI, currently underexplored in existing scholarship—such as informal data work or the territorial principle and its implications on AI law enforcement—this study investigates AI's critical vulnerabilities within the Latin American context in order to determine and provide regional and national policies to advance on an inclusive, strategic, and ethical approach for developing and deploying AI systems in Latin America. The study seeks to answer the question through a cross-analysis and comparative case study of six countries (Brazil, Chile, México Costa Rica, El Salvador and Honduras) drawing on existing and recent global and regional benchmarks, including the Stanford HAI AI Index (2024), UNESCO’s Recommendation on the Ethics of AI (2021), and the Latin American AI Index (ILIA 2024). The countries were selected based on a broad range of AI readiness levels, focusing on mapping institutional, regulatory, and socio-political contexts as well as metrics and input from relevant sources. The analysis shows structural inequality as the core vulnerability shaping AI’s impact in Latin America, alongside governance gaps, limited regional cooperation, and minimal public participation. The analysis identifies ten critical vulnerabilities—including the use of AI in surveillance, increase in inequality, increase in disinformation, AI-use in organized crime, and environmental exploitation—that, if unaddressed, may accelerate democratic erosion and technological dependency. Ethical principles are shown to be deeply interconnected and grounded in human rights, yet their implementation remains aspirational. This research underscores a call for action toward regional coordination, inclusive education strategies prioritizing gender policies and rural areas, and aligned industrial policies in the countries of the region. A Latin American context-specific, collective approach ensures that AI serves the public interest, strengthens sovereignty, and supports equitable development in Latin America.

Origin and Correlates of Viral Rebound in SIV-Infected Rhesus Macaques Following Discontinuation of ART

Thu, 01 May 2025 00:00:00 GMT

Origin and Correlates of Viral Rebound in SIV-Infected Rhesus Macaques Following Discontinuation of ART King, Irena V. The earliest events of viral rebound following discontinuation of ART in people living with Human Immunodeficiency Virus-1 remain largely unknown. We investigated detailed reservoir characteristics and viral rebound dynamics in 18 Simian Immunodeficiency Virus-infected rhesus macaques treated with antiretroviral therapy for 70 weeks and then necropsied after a 12-day analytical treatment interruption (ATI). Using molecularly barcoded SIVmac239M, we tracked viral clonotypes following ATI in both peripheral blood and tissues at necropsy. Viral rebound appeared to originate by reactivation of a single or a few barcode clonotypes from a limited number of deep lymph nodes or gastrointestinal tissues, followed by rapid virus replication of this clonotype in peripheral blood and tissues as well as serial reactivation of multiple additional barcode clonotypes from different anatomic sites, resulting in oligoclonal plasma viremia. Daily transcriptomic and proteomics profiling in peripheral blood following ATI identified early upregulation of pathways related to T cell signaling, cytokine responses, and metabolism prior to detectable plasma viremia, presumably reflecting initial viral replication in tissues. Taken together, these data provide a detailed anatomic, virologic, and immunologic characterization of viral rebound in SIV-infected macaques following ATI, which provides critical information to inform the development of next generation HIV-1 cure strategies.

Multiscale Modeling of Genome Organization: Bridging Polymer Physics, Molecular Dynamics, and AI

Thu, 01 May 2025 00:00:00 GMT

Multiscale Modeling of Genome Organization: Bridging Polymer Physics, Molecular Dynamics, and AI Lao, Zhuohan The human genome is intricately organized within the nucleus, and its spatial arrangement plays a critical role in gene regulation, cellular function, and disease. Recent advances in high- throughput experiments have unveiled the heterogeneous and dynamic nature of chromatin organization at single-cell resolution. However, computational tools that can both simulate and predict such complex structures are still limited. In this thesis, we develop and apply computational frameworks to investigate nuclear genome organization at high spatial and temporal resolution. Our approaches integrate biophysical modeling and generative artificial intelligence to address complementary aspects of nuclear architecture. In Chapter 1, we provide an overview of the hierarchical organization of the genome and discuss emerging principles that govern chromatin folding, nuclear compartmentalization, and their functional implications. We introduce data-driven, physics-based, and generative artificial intelligence modeling approaches, highlighting the need for interpretable and efficient models capable of capturing the structural diversity of the nucleus across individual cells. In Chapter 2, we present OpenNucleome, a high-resolution molecular dynamics framework for simulating the entire human nucleus at 100-kilobase resolution. OpenNucleome incorpo- rates explicit representations of chromosomes, nuclear bodies, and the nuclear lamina, and faithfully reproduces experimental data from Hi-C, TSA-seq, DamID, and DNA-MERFISH. The developed software is fully open-source and GPU-accelerated, enabling large-scale simu- lations and mechanistic explorations. In Chapter 3, we explore the impact of genome organization on various biological phe- nomena within the cell nucleus—focusing on telomere and telomere condensate dynamics, and nuclear deformation—using OpenNucleome. Our results demonstrate that the three- dimensional genome architecture plays a pivotal role in governing the dynamics of genomic loci such as telomeres, influencing the kinetics and outcomes of droplet coarsening. More- over, specific interactions between the genome and nuclear bodies form robustly across cells, providing strong support for a nuclear zoning model of genome function. In Chapter 4, we introduce ChromoGen, a generative diffusion model that predicts single- cell chromatin conformations de novo from DNA sequence and DNase-seq data. Unlike traditional simulation frameworks, ChromoGen learns from experimental single-cell 3D structures to generate physically realistic, region- and cell type-specific ensembles. ChromoGen achieves high agreement with both experimental Dip-C and Hi-C data while maintaining computational efficiency, enabling rapid exploration of chromatin heterogeneity across the genome and cell types. Together, these two frameworks—OpenNucleome and ChromoGen—provide powerful and complementary tools for understanding genome structure and function at the single-cell level, bridging physics-based modeling and deep generative artificial intelligence modeling.

On single-cell immune dynamics of chronic HIV infection and treatment in rhesus macaque models

Thu, 01 May 2025 00:00:00 GMT

On single-cell immune dynamics of chronic HIV infection and treatment in rhesus macaque models Quinn, Sarah Lynne Human Immunodeficiency Virus (HIV) continues to be an overwhelming challenge in both global health and immunology. With no cure available, 30 million people worldwide rely on antiretroviral therapy (ART) to prevent transmission and disease progression. However, individuals on ART are at elevated risk for numerous comorbidities and are susceptible to continued disease progression should treatment be stopped or interrupted. Addressing the challenges resulting from treatment and lack of cure requires a deeper understanding of the complex underlying immunology of HIV infection, treatment, and therapeutics. Single-cell RNA sequencing (scRNAseq) is continually advancing our understanding of immune dynamics and when combined with well-characterized rhesus macaque models of HIV, provides an opportunity to profile immune perturbations over extensive time courses in a controlled setting. In this thesis, I present two studies that further our understanding of the host immune response across stages of infection, treatment, and therapeutic intervention, using rhesus macaque models. In the first study (Chapter 2), I comprehensively profiled immune dynamics during untreated infection, ART initiation, and long-term ART, leveraging a longitudinal cohort of SIV-infected (Simian Immunodeficiency Virus) macaques. This work is particularly relevant given the increasing age and average time spend on ART among people living with HIV. scRNAseq revealed key immune shifts during acute and chronic infection, as well as over five years of subsequent ART. I identified cell type composition shifts during prolonged untreated infection and uncovered areas of unresolved immune dysregulation despite long-term ART– most prominently among myeloid gene expression and enrichment. I further link transcriptional changes to intact proviral burden and identified ribosomal pathways as markers infection stage, treatment status, reservoir size. Finally, by evaluating published immune correlates of treatment outcome, I identify nuances in signatures changing and remaining stable with time on ART. In the second study (Chapter 3), I expand on the baseline infection and treatment case by evaluating immune dynamics in response to a post-exposure combination therapeutic (Ad26/MVA + PGT121 + Vesatolimod) previously shown to induce post-ART viral control in most (7/10) treated macaques with Simian Human Immunodeficiency Virus (SHIV). Here I identify features of therapeutic response and implicate a previously defined Antibody-Dependent Cellular Phagocytosis signature as being associated with control. Furthermore, I identify a new cytotoxic transcriptional module in T and NK cells associated with both non-rebounding animals and post-rebound controller animals, suggesting a shared effector program associated with successful virologic control. Supported by a thorough introduction on immunological techniques, questions and applications to HIV studies (Chapter 1), and a discussion of intersectionality and future directions of the field (Chapter 4), this thesis provides a comprehensive analysis of immune dynamics across the lifecycle of viral infection, treatment, and therapeutic intervention in macaque models of HIV. This work demonstrates how the host immune environment influences therapeutic success, laying a foundation for future therapeutic design.

Adsorption and electrostatic potentials at the electrochemical interface

Thu, 01 May 2025 00:00:00 GMT

Adsorption and electrostatic potentials at the electrochemical interface Nowack, Linsey This paper explores adsorption in electrochemical systems. Part I reviews ways to quantify adsorption, the energetic considerations that make adsorption favorable or unfavorable, how to measure adsorption experimentally, and how it has previously been modeled using extensions of the Langmuir isotherm. At the end of Part I, a simple Monte Carlo model (MC) is applied to a very complex carbon dioxide reduction system to study competitive adsorption. This application of Monte Carlo simulations demonstrates the challenges of extracting meaningful parameters from empirically fitting MC simulations to isotherms derived from nanoparticle-enhanced Raman spectra. Part II examines how adsorption influences the electrostatic potential in the electrochemical double layer using molecular dynamics simulations. Adding on to previous work from the Willard group, this chapter calculates how adsorbate polarity and coverage influences two characterizations of Coulombic interactions: the Poisson potential and the Madelung potential. Both potentials, while having different shapes as a function of distance from the electrode surface, exhibit strong sensitivity to water structure. At high coverage, adsorbates decrease the number of interfacial waters, shifts the position of the molecular layers of waters at the interface, and disrupts the water's orientational order. Lastly, cross-sections of the 3D Poisson potential parallel to the electrode surface reveal large heterogeneity in Poisson potential values as a result of adsorbates. This suggests that 1D electrostatic potential profiles are not enough to understand forces in the electrochemical double layer.

The combustion of droplets of heavy liquid fuels

Fri, 01 Jan 1954 00:00:00 GMT

The combustion of droplets of heavy liquid fuels Simpson, Hugh Cameron. Thesis: Sc. D., Massachusetts Institute of Technology, Department of Chemical Engineering, 1954; Vita.; Bibliography: leaves 542-552.

The melting points of isopropyl esters of aromatic nitro acids

Mon, 01 Jan 1923 00:00:00 GMT

The melting points of isopropyl esters of aromatic nitro acids Zeng, Zhaolun, 1898-1967. Thesis: B.S., Massachusetts Institute of Technology, Department of Chemical Engineering, 1923

On the use of selected derivatives in the specific characterization of alcohols, phenols, amines and mercaptans

Fri, 01 Jan 1926 00:00:00 GMT

On the use of selected derivatives in the specific characterization of alcohols, phenols, amines and mercaptans Zeng, Zhaolun, 1898-1967. Thesis: Sc. D., Massachusetts Institute of Technology, Department of Chemistry, 1926; Vita.

An investigation of certain concrete sands in the vicinity of Portland, Maine

Sat, 01 Jan 1927 00:00:00 GMT

An investigation of certain concrete sands in the vicinity of Portland, Maine Blandford, Sidney E. (Sidney Edgar); Cheney, Laurence B. Thesis: B.S., Massachusetts Institute of Technology, Department of Civil Engineering, 1927

Measurements on contact potentials of metals

Sun, 01 Jan 1928 00:00:00 GMT

Measurements on contact potentials of metals Zisman, William A. Thesis: M.S., Massachusetts Institute of Technology, Department of Physics, 1928; Includes bibliographical references (leaves 59-60).

A department store fore the Hudson Bay Company Edmundton, Alberta, Canada

Sat, 01 Jan 1938 00:00:00 GMT

A department store fore the Hudson Bay Company Edmundton, Alberta, Canada Thrift, Eric W. Thesis: M. Arch., Massachusetts Institute of Technology, Department of Architecture, 1938

The misfit and the non-specialist student at M.I.T.

Sun, 01 Jan 1939 00:00:00 GMT

The misfit and the non-specialist student at M.I.T. Peskoe, Irving. Thesis: B.S., Massachusetts Institute of Technology, Department of General Science, 1939; Includes bibliographical references (leaf 64).

The buckling load of hyperbolic paraboloid shells

Mon, 01 Jan 1962 00:00:00 GMT

The buckling load of hyperbolic paraboloid shells Lee, Samuel Tak. Thesis: B.S., Massachusetts Institute of Technology, Department of Civil and Sanitary Engineering, 1962; Includes bibliographical references (leaf 85).

A study of maladjustment in the class of 1955 at M. I. T.

Sat, 01 Jan 1955 00:00:00 GMT

A study of maladjustment in the class of 1955 at M. I. T. Langberg, Arnold. Thesis: B.S., Massachusetts Institute of Technology, Department of General Engineering, 1955

Education in the humanities at the Massachusetts Institute of Technology.

Sat, 01 Jan 1966 00:00:00 GMT

Education in the humanities at the Massachusetts Institute of Technology. Perrolle, Judith Ann. Thesis: B.S., Massachusetts Institute of Technology, Department of Humanities, 1966; Bibliography: leaves 98-99.

Going to MIT.

Fri, 01 Jan 1965 00:00:00 GMT

Going to MIT. Landau, David Lewis. Thesis: B.S., Massachusetts Institute of Technology, Department of Humanities, 1965

The chemistry of octahalodirhenate (III).

Sat, 01 Jan 1966 00:00:00 GMT

The chemistry of octahalodirhenate (III). Robinson, William Robert. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Chemistry, 1966; Bibliography: leaves 112-116.

The 1970 student strike: the children's crusade at MIT; a study in contemporary history.

Mon, 01 Jan 1973 00:00:00 GMT

The 1970 student strike: the children's crusade at MIT; a study in contemporary history. Giguere, Lee David. Thesis: B.S., Massachusetts Institute of Technology, Department of Humanities, 1973; Leaf 69 used twice.; Bibliography: leaf 163.

The style of M.I.T. education.

Thu, 01 Jan 1970 00:00:00 GMT

The style of M.I.T. education. Green, Patrick Conal. Thesis: B.S., Massachusetts Institute of Technology, Department of Humanities, 1970; Bibliography: leaf 55.

On the economics of advertising.

Thu, 01 Jan 1970 00:00:00 GMT

On the economics of advertising. Schmalensee, Richard. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Economics, 1970; Vita.; Bibliography: leaves 485-496.

An essay on taxation and growth in an enclave economy.

Mon, 01 Jan 1968 00:00:00 GMT

An essay on taxation and growth in an enclave economy. Francis, Alfred Alexander Jaques. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Economics, 1968; Bibliography: leaves 130-131.

An examination of low hysteresis polyurethane for use in traction drive rolling elements

Thu, 01 Jan 1981 00:00:00 GMT

An examination of low hysteresis polyurethane for use in traction drive rolling elements Castellano, John Philip. Thesis: B.S., Massachusetts Institute of Technology, Department of Mechanical Engineering, 1981; Includes bibliographical references.

Career goals, attitudes, and interpersonal relationships of M.I.T. undergraduates

Fri, 01 Jan 1982 00:00:00 GMT

Career goals, attitudes, and interpersonal relationships of M.I.T. undergraduates Slaughter, Sarah. Thesis: B.S., Massachusetts Institute of Technology, Department of Humanities, 1982; Bibliography: leaf 99.

The work and meanings of science and engineering : interviews with MIT scholars

Fri, 01 Jan 1982 00:00:00 GMT

The work and meanings of science and engineering : interviews with MIT scholars Karaku, Alex Theodore. Thesis: B.S., Massachusetts Institute of Technology, Department of Humanities, 1982; Bibliography: leaf 14.

Nonpolarized relay

Fri, 01 Jan 1915 00:00:00 GMT

Nonpolarized relay Tseng, C. C. Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering, 1915

The Hiddenness Argument and The Limits of Doxastic Positioning

Sat, 01 Feb 2025 00:00:00 GMT

The Hiddenness Argument and The Limits of Doxastic Positioning Garcia, Nicole If God exists, how clear or obvious should we expect his existence to be? Particularly if such a God is interested in having a personal relationship with us? The Hiddenness Argument contends that it should be much clearer than it in fact is. If God exists and really wants us to know as much, we should expect to inhabit a very different epistemic situation than we in fact do – one that rules out the possibility of rational nonbelief. The evidence available for God’s existence should be so definitive that it would be impossible for us to fail to believe on good epistemic terms. My dissertation sets out to delegitimize this expectation. While available objections to it challenge its propriety – God may have overriding reasons for disclosing his existence in a way that allows for rational nonbelief – my account challenges its feasibility – whether it is in principle possible to meet. Expecting divine self-disclosure to rule out rational nonbelief assumes that it can rule out rational nonbelief – but can it? By homing in on the nature, mechanics, and limitations of disclosure itself, I show it cannot. Divine self-disclosure is an instance of what I call doxastic positioning: the process of positioning someone to rationally form some belief – in this case, theistic belief. To rule out rational nonbelief, God’s disclosure would need to make theistic belief a universal rational requirement. Given the success conditions of doxastic positioning, this would involve the provision of sufficient evidence as well as the universal possession and appreciability of said evidence. But no matter what evidence God provides, God cannot guarantee on pain of irrationality that humans will possess or be in a position to appreciate the available evidence, leaving rational nonbelief an ever live possibility. —Which is just to say that divine self-disclosure cannot rule out rational nonbelief. The expectation that it would do so is, then, illegitimate and the hiddenness argument depending on it fails.

Proteolethargy is a pathogenic mechanism in chronic disease

Thu, 01 May 2025 00:00:00 GMT

Proteolethargy is a pathogenic mechanism in chronic disease Moreno, Shannon The pathogenic mechanisms of many diseases are well understood at the molecular level, but there are prevalent syndromes associated with pathogenic signaling, such as diabetes and chronic inflammation, where our understanding is more limited. Here, I present evidence that pathogenic signaling suppresses the mobility of a spectrum of proteins that play essential roles in cellular functions known to be dysregulated in these chronic diseases. The reduced protein mobility, which we call proteolethargy, was linked to cysteine residues in the affected proteins and signaling-related increases in excess reactive oxygen species. Diverse pathogenic stimuli, including hyperglycemia, dyslipidemia, and inflammation, produce similar reduced protein mobility phenotypes. I propose that proteolethargy is an overlooked cellular mechanism that may account for various pathogenic features of diverse chronic diseases.

Neural Sequences Underlying Directed Turning in C. elegans

Thu, 01 May 2025 00:00:00 GMT

Neural Sequences Underlying Directed Turning in C. elegans Kramer, Talya Complex behaviors like navigation rely on sequenced motor outputs that combine to generate effective movement. The brain-wide organization of the circuits that integrate sensory signals to select and execute appropriate motor sequences is not well understood. Here, we characterize the architecture of neural circuits that control C. elegans olfactory navigation. We identify error-correcting turns during navigation and use whole-brain calcium imaging and cell-specific perturbations to determine their neural underpinnings. These turns occur as motor sequences accompanied by neural sequences, in which defined neurons activate in a stereotyped order during each turn. Distinct neurons in this sequence respond to sensory cues, anticipate upcoming turn directions, and drive movement, linking key features of this sensorimotor behavior across time. The neuromodulator tyramine coordinates these sequential brain dynamics. Our results illustrate how neuromodulation can act on a defined neural architecture to generate sequential patterns of activity that link sensory cues to motor actions.

High-Dimensional Statistics for Causal Inference and Panel Data

Thu, 01 May 2025 00:00:00 GMT

High-Dimensional Statistics for Causal Inference and Panel Data Klosin, Sylvia This dissertation develops new econometric tools for causal inference in panel data settings, with a focus on addressing key biases that arise in high-dimensional and dynamic environments. While this dissertation is motivated by the need to flexibly measure the economic impacts of climate change, the methods I develop are much more general. They apply broadly to panel data problems across empirical economics—including in labor, development, and industrial organization—where standard fixed effects estimators may fail. The first chapter identifies a previously overlooked source of bias in fixed effects panel estimators, which I term dynamic bias. This bias arises when dynamic feedback—where past outcomes influence current outcomes—is ignored in the estimating equation. I show that dynamic bias can be severe even when treatments are randomly assigned and that it often exceeds the well-known Nickell bias. To address this, I develop a bias-corrected estimator that is consistent in panels with a fixed number of time periods. I apply this method to estimate the effects of temperature shocks on GDP, where accounting for dynamic feedback reduces estimated damages substantially. The second chapter, coauthored with Max Vilgalys, proposes a flexible estimator for continuous treatment effects using panel data with fixed effects. We extend the double debiased machine learning (DML) framework to this setting and prove consistency and asymptotic normality. In an application to U.S. agriculture, we show that our estimator captures nonlinear effects of temperature on crop yields more accurately than standard linear models, estimating substantially larger damages from extreme heat. The final chapter further generalizes the methodological contribution by introducing a non-parametric estimator of the average dose-response function. Building on recent developments in DML and automatic double machine learning (ADML), I propose a novel debiasing strategy that directly estimates the bias correction term, yielding favorable theoretical properties. Together, these essays provide practical and theoretically grounded tools for applied researchers working with panel data, particularly in settings characterized by high dimensionality, continuous treatments, or dynamic feedback.

Essays in Health Economics

Thu, 01 May 2025 00:00:00 GMT

Essays in Health Economics Moran, Kelsey C. This dissertation comprises three essays in health economics. The first paper studies how imperfect electronic health record (EHR) system compatibility, or interoperability, affects patients. My coauthors, Rebekah Dix and Thi Mai Anh Nguyen, and I find that improved EHR interoperability between hospitals leads to better health outcomes and lower costs for shared patients. We also show that hospitals prefer sending patients to facilities with more compatible EHR systems, causing patient reallocation across providers based on technological factors. Using a model of patient flows, we estimate that eliminating these frictions would generate substantial welfare gains by improving patient outcomes and reducing allocative distortions. The second paper examines how regulatory requirements influence hospital charity care by analyzing the Hill-Burton Act of 1946, which allocated $6 billion to over 3,500 hospitals in exchange for providing free care to uninsured patients. I find that after these obligations expire, hospitals strategically reduce charity care by 30% and decrease admissions of charity-eligible patients by 14%. These patients subsequently shift to neighboring public and non-profit hospitals, where they must pay for care and experience higher mortality rates. The third paper, co-authored with Ari Bronsoler, Joseph Doyle, and John Van Reenen, studies the broad impact of Health Information Exchange (HIE) on patient outcomes. Using a newly compiled database of state HIE laws as instruments for hospital HIE, we find that HIE significantly reduces mortality from infectious diseases and hospital readmission rates for common conditions. With HIE usage increasing by 50 percentage points from 2009 to 2019, we estimate this technology saved approximately 27,000 lives annually through improved care coordination and public health response.

Essays on Firms and Technology in Development Economics

Thu, 01 May 2025 00:00:00 GMT

Essays on Firms and Technology in Development Economics Houeix, Deivy My thesis investigates the relationships between technology and firms in lower-income countries. I explore both the economic impacts of technology on firms—how it affects their economic outcomes and their relationships with stakeholders, both within and across firms—and the determinants of technology adoption: what underlying factors impede or drive the uptake of new technologies? I combine a diverse set of methods, including large-scale field experiments, economic theory, and long-term collaborations with local partners to investigate how small firms—such as taxis, retailers, and others—transform their practices and relationships as they adopt new technologies. My work centers on West Africa, one of the world’s poorest regions, where economic research remains limited. Chapter 1: The first chapter investigates the idea that digital technologies have the potential to increase firm productivity. However, they often come bundled with data observability, which can be a double-edged sword. Observability reduces information frictions and can increase efficiency, but some agents may lose their informational rent and thus resist adoption. I explore this trade-off between observability and adoption through two field experiments conducted over nearly two years. These experiments, guided by contract theory, introduce digital payments to the Senegalese taxi industry in partnership with the country's largest payment company. In the first experiment, I randomize access to digital payments for drivers (employees) and transaction observability to taxi owners (employers). I find that digital payments reduce drivers' cash-related costs by about half but also serve as effective monitoring tools for taxi owners. Transaction observability substantially increases driver effort, contract efficiency, and the duration of owner-driver relationships. However, 50% of drivers—primarily the worst-performing and poorest—decline to adopt digital payments when transactions are observable. The second experiment shows that the adoption rate doubles when drivers are assured that owners will not be able to observe their transactions. I develop a theoretical framework and use the experimental variations to estimate the welfare impacts of policy counterfactuals. I show that removing transaction observability would maintain moral hazard problems but broaden adoption and thus increase overall welfare—an approach ultimately implemented by the payment company. These findings highlight the crucial role of information embedded in digital technologies, as it magnifies gains for adopting firms but can deter initial adoption. Chapter 2: In the second chapter, I conduct a randomized experiment to study the nationwide technology diffusion of a new digital payment technology in Senegal. By leveraging two novel sources of network data—mobile money transactions and anonymized phone contact directories covering the near universe of the adult population in Senegal—I causally identify three sets of adoption spillovers from taxi firms randomized to receive early access to the technology: intra-industry among taxi firms; inter-industry between taxi drivers and other small businesses; and inter-regional spillovers from the capital city to businesses in other urban centers. I show that spillovers go beyond strategic complementarities, reflecting social learning within firms' social networks, driven by social ties and remote interactions. Chapter 3: In the third and final chapter, I explore the fact that search and trust frictions have historically made it hard for small firms in lower-income countries to buy inputs from foreign markets. The growth in smartphone ownership and social media usage has the potential to alleviate these barriers. Informed by a dynamic model of relational contracting, we run a field experiment leveraging these technological tools to provide exogenous variation in (1) search frictions and (2) trust frictions (adverse selection and moral hazard) in a large international import market. In our search treatment, we connect a randomly selected 80% of 1,862 small garment firms in Senegal to new suppliers in Turkey. We then cross-randomize two trust treatments that provide additional information about the types (adverse selection) and incentives (moral hazard) of these new suppliers. Alleviating search frictions is sufficient to increase access to foreign markets: in all treated groups, firms are 26% more likely to have the varieties a mystery shopper requests and the goods sold are 30% more likely to be high quality. However, the trust treatments are necessary for longer-term impact: using both transaction-level mobile payments data and a follow-up survey, we show that these groups are significantly more likely to develop the connections into relationships that persist beyond the study. These new relationships lead to increases in medium-run profit and sales. Finally, we use the treatment effects to estimate the model and evaluate counterfactuals where we set various combinations of the frictions to zero, finding that the largest gains come from eliminating adverse selection.

Essays in Industrial Policy, Misallocation and Production Networks

Thu, 01 May 2025 00:00:00 GMT

Essays in Industrial Policy, Misallocation and Production Networks Garg, Tishara The thesis comprises three chapters studying industrial policy, misallocation and macroeconomic propagation in developing countries. The first chapter studies the impact of placebased industrial policies on equilibrium selection with Indian industrial parks as the empirical context. The second and third chapters study firm networks in Chile and Turkey respectively— using a common theoretical framework, the former studies the incidence of distortions while the latter studies the propagation of a large refugee shock. The first chapter introduces a method to study the impact of policy events on equilibrium selection in settings where strong complementarities may lead to multiple equilibria and coordination failures. Many industrial policies are rooted in the idea of coordination failures and big-push’ theories, yet empirical evidence on their effectiveness remains limited, since distinguishing equilibrium shifts from direct changes in fundamentals is challenging. Leveraging tools from industrial organization and algebraic geometry, I develop an approach to study coordination effects without imposing strong assumptions on the distribution or responsiveness of economic fundamentals. The method identifies the ‘types’ of factual and counterfactual equilibria through a three-step procedure: model estimation and inversion, equilibrium enumeration, and type assignment. Types of factual equilibria may be used to examine how events, like urban infrastructure, subsidy drives, or trade liberalization, affect equilibrium selection. Types of counterfactual equilibria further allow decomposition of observed effects into fundamentals- versus coordination-driven. I apply this method to study industrial zones in India. Using a newly assembled dataset, I find that municipalities receiving an industrial zone see a 60% increase in non-farm employment over 15 years, with significant spillovers to non-targeted sectors and municipalities. Combining the methodology with event study designs, I find that industrial zones increase the probability of escaping a low-industrialization equilibrium by 38%, with coordination effects explaining roughly onethird of the observed change in outcomes. The second chapter (joint with David Atkin, Baptiste Bernadac, Dave Donaldson, and Federico Huneeus) combines unique datasets from Chile to quantify the full incidence of distortions for the first time. Economic distortions—such as market power, taxes, credit constraints, etc.—are fundamental in understanding the difference between developing and developed economies. Recent work has documented the pervasive extent of economic distortions and how they lead to substantial misallocation, or aggregate productivity loss. Far less well understood is how these phenomena affect members of society differently. We embed a new dataset which we build by linking workers and owners to firms, firms to each other, firms to consumers, and firms and consumers to the government, inside a general equilibrium model of the Chilean economy. Armed with internal estimates of distortions on exchanges throughout the economy, as well as data on the network of such linkages, we conduct a series of counterfactual simulations that illuminate the incidence of distortions in our model economy. We find that the burden of distortions falls relatively more on the shoulders of the poor, the young and women. The final chapter (joint with Ahmet Gulek) investigates how immigration-induced wage shocks can propagate beyond the regions receiving immigrants through the production network. Using the Syrian refugee crisis in Turkey as a quasi-experiment and the near universe of domestic firm-to-firm transaction data from VAT records, we show that the immigration shock propagates both forward and backward along the supply chain. Firms in non-host regions who directly or indirectly buy from host regions demand more labor. Firms who sell to host regions weakly increase their sales. Estimates imply an elasticity of substitution between labor and intermediate goods of 0.76 and an elasticity of substitution of nearly 1 between intermediates. Counterfactual analyses show that the spillover effects on non-host regions are economically meaningful when the host regions are central nodes of the domestic trade network. For example, a 1% increase in labor supply in Istanbul decreases real wages in Istanbul by 0.56% and increases real wages in the average non-host city by 0.38%.