Geometric Properties of Learned Representations
In machine learning, reprensentation learning refers to optimizing a mapping from data to some representation space (usually generic vectors in Rᵈ for some pre-determined 𝑑 much lower than data dimensions). While such training often uses no supervised labels, the learned representations have proved...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2023
|
| Online Access: | https://hdl.handle.net/1721.1/147353 |
| _version_ | 1826203741268738048 |
|---|---|
| author | Wang, Tongzhou |
| author2 | Isola, Phillip |
| author_facet | Isola, Phillip Wang, Tongzhou |
| author_sort | Wang, Tongzhou |
| collection | MIT |
| description | In machine learning, reprensentation learning refers to optimizing a mapping from data to some representation space (usually generic vectors in Rᵈ for some pre-determined 𝑑 much lower than data dimensions). While such training often uses no supervised labels, the learned representations have proved very useful for solving downstream tasks. Such successes sparkled an enormous amount of interests in representation learning methods among both academic researchers and practitioners. Despite the popularity, it is not always clear what the representation learning objectives are optimizing for, and how to design representation learning methods for new domains and tasks (such as reinforcement learning). In this thesis, we consider the structures captured by two geometric properties of learned representations: invariances and distances. From these two perspectives, we start by thoroughly analyzing the widely adopted contrastive representation learning, uncovering that it learns certain structures and relations among data. Then, we describe two new representation learning methods for reinforcement learning and control, where they respectively capture the optimal planning cost (distance) and the information invariant to environment noises. |
| first_indexed | 2024-09-23T12:42:19Z |
| format | Thesis |
| id | mit-1721.1/147353 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T12:42:19Z |
| publishDate | 2023 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/1473532023-01-20T03:32:24Z Geometric Properties of Learned Representations Wang, Tongzhou Isola, Phillip Torralba, Antonio Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science In machine learning, reprensentation learning refers to optimizing a mapping from data to some representation space (usually generic vectors in Rᵈ for some pre-determined 𝑑 much lower than data dimensions). While such training often uses no supervised labels, the learned representations have proved very useful for solving downstream tasks. Such successes sparkled an enormous amount of interests in representation learning methods among both academic researchers and practitioners. Despite the popularity, it is not always clear what the representation learning objectives are optimizing for, and how to design representation learning methods for new domains and tasks (such as reinforcement learning). In this thesis, we consider the structures captured by two geometric properties of learned representations: invariances and distances. From these two perspectives, we start by thoroughly analyzing the widely adopted contrastive representation learning, uncovering that it learns certain structures and relations among data. Then, we describe two new representation learning methods for reinforcement learning and control, where they respectively capture the optimal planning cost (distance) and the information invariant to environment noises. S.M. 2023-01-19T18:47:38Z 2023-01-19T18:47:38Z 2022-09 2022-10-19T18:58:58.269Z Thesis https://hdl.handle.net/1721.1/147353 0000-0002-0559-9775 In Copyright - Educational Use Permitted Copyright MIT http://rightsstatements.org/page/InC-EDU/1.0/ application/pdf Massachusetts Institute of Technology |
| spellingShingle | Wang, Tongzhou Geometric Properties of Learned Representations |
| title | Geometric Properties of Learned Representations |
| title_full | Geometric Properties of Learned Representations |
| title_fullStr | Geometric Properties of Learned Representations |
| title_full_unstemmed | Geometric Properties of Learned Representations |
| title_short | Geometric Properties of Learned Representations |
| title_sort | geometric properties of learned representations |
| url | https://hdl.handle.net/1721.1/147353 |
| work_keys_str_mv | AT wangtongzhou geometricpropertiesoflearnedrepresentations |