Abstract
dc:description.abstractOptimal transport (OT) distances provide a way to compare probability measures with rich geometric and topological properties that align well with human perception in many areas of machine learning. However, OT distances suffer from computational and statistical scalability issues to high dimensions, which motivated the study of regularized OT methods like slicing, smoothing, and entropic penalty. In this thesis, I will discuss several applications of regularized OT distances towards problems in non-parametric inference and generative modelling, and how regularization helps address some issues with vanilla OT. I will also discuss some aspects of dimension adaptation of regularized OT distances, and the unregularized semi-discrete setting where the statistical convergence issue can also be resolved.
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
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Sadhu, Ritwik
- Chair dc:contributor.chair
-
- Kato, Kengo
- Committee members dc:contributor.committeemember
-
- El Alaoui El Abidi, Ahmed
- Goldfeld, Ziv
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Attribution 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 16611774
- Dc Identifier Other
-
ProQuest Submission ID: 14434
ProQuest Publication ID: 31337288