Electrical Engineering and Computer Sciences - Master's degree
http://hdl.handle.net/1721.1/7663
2014-10-25T01:36:45ZNeural network architectures for Prepositional Phrase attachment disambiguation
http://hdl.handle.net/1721.1/91147
Neural network architectures for Prepositional Phrase attachment disambiguation
Belinkov, Yonatan
This thesis addresses the problem of Prepositional Phrase (PP) attachment disambiguation, a key challenge in syntactic parsing. In natural language sentences, a PP may often be attached to several possible candidates. While humans can usually identify the correct candidate successfully, syntactic parsers are known to have high error rated on this kind of construction. This work explores the use of compositional models of meaning in choosing the correct attachment location. The compositional model is defined using a recursive neural network. Word vector representations are obtained from large amounts of raw text and fed into the neural network. The vectors are first forward propagated up the network in order to create a composite representation, which is used to score all possible candidates. In training, errors are back-propagated down the network such that the composition matrix is updated from the supervised data. Several possible neural architectures are designed and experimentally tested in both English and Arabic data sets. As a comparative system, we offer a learning-to-rank algorithm based on an SVM classifier which has access to a wide range of features. The performance of this system is compared to the compositional models.
Thesis: S.M. in Computer Science and Engineering, Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2014.; This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.; 25; Cataloged from student-submitted PDF version of thesis.; Includes bibliographical references (pages 41-44).
2014-01-01T00:00:00ZA delay-constrained cross-layer model using network coding
http://hdl.handle.net/1721.1/91099
A delay-constrained cross-layer model using network coding
Adams, David C. (David Christopher)
Traditionally, most packet-switched networks have only one wireless hop: the link between the end users and their access point. However, there is increasing interest in using wireless links to reach the edge of the network. Having more than one wireless link is a game changer. Network layer architecture is predicated on the assumption that the lower layers are reliable, but this comes at a high cost in terms of data rate on a band-limited, lossy wireless channel. This cost is tolerable over one underutilized link, but when the network demands high-capacity wireless links, it may be time to rethink the way the packet-switched network interacts with its underlying infrastructure. The aim of this thesis is to provide a general model that can be used to frame a wide variety of cross-layer coding problems. We do not explicitly consider the channel code, medium access, or modulation; instead, we leverage the maturity of these fields to observe the general effect they produce on higher layers. We focus our model on applications where delay is constrained, which forces us to consider coding results in the regime where code length is non-asymptotically large. In trying to extend our analysis to multi-hop flows, we develop an analytical tool that can be useful in wider applications. This tool simplifies certain network flows to a distribution on the amount of information available to the destination; it is a step towards characterizing network information flows in the non-asymptotic regime. Finally, we will use the model to design encoding schemes, given practically-motivated constraints.
Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2014.; 27; Cataloged from PDF version of thesis.; Includes bibliographical references (pages 79-81).
2014-01-01T00:00:00ZBetter embeddings for Planar Earth-Mover Distance over sparse sets
http://hdl.handle.net/1721.1/91098
Better embeddings for Planar Earth-Mover Distance over sparse sets
Backurs, Arturs
We consider the problem of constructing low-distortion embeddings of the Planar Earth-Mover Distance (EMD) into lp spaces. EMD is a popular measure of dissimilarity between sets of points, e.g., bags of geometric features. We present a collection of embeddings with the property that their distortion and/or host-space dimension are parametrized by the size (or the sparsity) of the embedded sets s. Our specific results include: -- An O(log s)-distortion embedding of EMD over s-subsets into l1-e. This is the first embedding of EMD into a "tractable" lp, space whose distortion is a function of the sparsity, not the size of the ambient space; -- An O(log n)-distortion embedding of EMD into lp, with dimension O(s2 log2 n), where the embedded sets are subsets of an n x n grid. For low values of s this significantly improves over the best previous dimension bound of 0(n 2 ) obtained for general sets.
Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2014.; 11; Cataloged from PDF version of thesis.; Includes bibliographical references (pages 43-44).
2014-01-01T00:00:00ZThe structure of promises in quantum speedups
http://hdl.handle.net/1721.1/91097
The structure of promises in quantum speedups
Ben David, Shalev
It has long been known that in the usual black-box model, one cannot get super-polynomial quantum speedups without some promise on the inputs. In this thesis, we examine certain types of symmetric promises, and show that they also cannot give rise to super-polynomial quantum speedups. We conclude that exponential quantum speedups only occur given "structured" promises on the input. Specifically, we show that there is a polynomial relationship of degree 12 between D(f) and Q(f) for any function f defined on permutations (elements of {0, 1, ... , M - 1}1 in which each alphabet element occurs exactly once). We generalize this result to all functions f defined on orbits of the symmetric group action S., (which acts on an element of {0, 1, . . . , M - I}f by permuting its entries). We also show that when M is constant, any function f defined on a "symmetric set" - one invariant under Sn - satisfies R(f) = O(Q(f)12(M-1)).
Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2014.; Cataloged from PDF version of thesis.; Includes bibliographical references (page 35).
2014-01-01T00:00:00Z