Abstract
dc:description.abstractRandom projections is a technique used primarily in dimension reduction, in order to estimate distances in data. They can be thought of a linear transformation mapping a data matrix X to a lower dimensional space, where distances are preserved in expectation. However, the preservation of distances can be thought of a stepping stone to some eventual goal, such as classification, hypothesis testing, information retrieval, or even reconstructing principal components of data. In this thesis, I will give a background of the basic random projection algorithm. Next, I then look at the structure of random projection matrices and propose modifications to result in a more accurate estimation of distances, which would help in information retrieval and reconstruction of principal components. Finally, I show that it is possible to juxtapose the use of Monte Carlo variance reduction methods with random projections to improve the accuracy of distance estimates, which can then be used in an algorithm or procedure of the users' choice. Theoretical justifications are given, and empirical results are shown with synthetic data, and experiments from publicly available datasets.
Degree
thesis:*- Name thesis:degree_name
- Ph. D., Statistics
- Level thesis:degree_level
- Doctor of Philosophy
- Discipline thesis:degree_discipline
- Statistics
- Grantor
- Cornell University
- Year dc:date.issued
- 2017
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kang, Keegan
- Chair dc:contributor.chair
-
- Hooker, Giles J
- Committee members dc:contributor.committeemember
-
- Sridharan, Karthik
- Mimno, David
Subjects
dc:subject × 5Rights
dc:rights- Statement dc:rights
-
- Attribution-ShareAlike 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 9948798
- Dc Identifier Other
-
ProQuest Submission ID: 10216
ProQuest Publication ID: 10260644