2026-09-18T19:52:19Zhttps://dspace.mit.edu/server/oai/request

oai:dspace.mit.edu:1721.1/1515122023-08-01T03:17:57Zcom_1721.1_7582com_1721.1_7581col_1721.1_131022

Chernozhukov, Victor Mikusheva, Anna Vijaykumar, Suhas Massachusetts Institute of Technology. Department of Economics 2023-07-31T19:45:23Z 2023-07-31T19:45:23Z 2023-06 2023-06-01T16:03:39.420Z https://hdl.handle.net/1721.1/151512 https://orcid.org/0000-0001-8383-5617 The thesis consists of three essays. The first, titled “Localization, Convexity, and Star Aggregation,” develops new analytical tools based upon the offset Rademacher complexity for studying stochastic optimization in non-convex domains, including statistical prediction and model aggregation problems. Using these tools, I show that a simple procedure called the star algorithm can recover near-optimal convergence rates for non-parametric logistic regression in non-convex models. The second essay, titled “Kernel Ridge Regression Inference,” introduces a new technique for deriving sharp, non-asymptotic, uniform Gaussian approximation for partial sums in a reproducing kernel Hilbert space, which is then applied to construct uniform confidence bands for the widely-used kernel ridge regression algorithm. The third and final essay, titled “Frank-Wolfe Meets Metric Entropy,” uses ideas from asymptotic geometry to derive new dimension-dependent and domain-specific lower bounds for conditional gradient algorithms, a class of optimization procedures including the popular Frank-Wolfe algorithm and many of its variants. Such algorithms have found extensive use in machine learning and high-dimensional statistics, motivating a more thorough analysis of their limitations in high-dimensional problems. Ph.D. Massachusetts Institute of Technology Attribution 4.0 International (CC BY 4.0) Copyright retained by author(s) https://creativecommons.org/licenses/by/4.0/ Essays on Algorithmic Learning and Uncertainty Quantification Thesis application/pdf Doctoral Doctor of Philosophy Publication unknown unknown <Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="c2c4c262-21a2-4686-a5df-f66edd345fdc"> <Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type> <Title>Essays on Algorithmic Learning and Uncertainty Quantification</Title> <PublishedIn> <Publication> </Publication> </PublishedIn> <PublicationDate>2023-06</PublicationDate> <Authors> <Author> <DisplayName>Vijaykumar, Suhas</DisplayName> <Affiliation> <OrgUnit> </OrgUnit> </Affiliation> </Author> </Authors> <Editors> </Editors> <Publishers> <Publisher> <DisplayName>Massachusetts Institute of Technology</DisplayName> <OrgUnit /> </Publisher> </Publishers> <License>https://creativecommons.org/licenses/by/4.0/</License> <Abstract>The thesis consists of three essays. The first, titled “Localization, Convexity, and Star Aggregation,” develops new analytical tools based upon the offset Rademacher complexity for studying stochastic optimization in non-convex domains, including statistical prediction and model aggregation problems. Using these tools, I show that a simple procedure called the star algorithm can recover near-optimal convergence rates for non-parametric logistic regression in non-convex models. The second essay, titled “Kernel Ridge Regression Inference,” introduces a new technique for deriving sharp, non-asymptotic, uniform Gaussian approximation for partial sums in a reproducing kernel Hilbert space, which is then applied to construct uniform confidence bands for the widely-used kernel ridge regression algorithm. The third and final essay, titled “Frank-Wolfe Meets Metric Entropy,” uses ideas from asymptotic geometry to derive new dimension-dependent and domain-specific lower bounds for conditional gradient algorithms, a class of optimization procedures including the popular Frank-Wolfe algorithm and many of its variants. Such algorithms have found extensive use in machine learning and high-dimensional statistics, motivating a more thorough analysis of their limitations in high-dimensional problems.</Abstract> <Access xmlns="http://purl.org/coar/access_right" > </Access> </Publication>