Deep representations with learned constraints
Abstract/Contents
- Abstract
- Learning to extract the task-relevant features from high-dimensional data is an important challenge in machine learning. The recent success of machine learning is largely attributable to the advancement of deep neural networks, which transform the data into a new representation that is amenable to downstream machine learning algorithms. Deep neural networks thus treat the extraction of task-relevant features as a representation learning problem to be solved jointly with the task of interest via end-to-end training. This dissertation dives deeper into the process of representation learning and argues that it is often possible to supplement the existing training signal by imposing additional constraints on the learned representations. These constraints allow us to imbue the representation space with desirable traits that are known a priori to benefit the downstream task of interest. In this dissertation, we shall consider examples of various tasks and then show how to leverage insights about the task to constrain the representation in a beneficial manner. These insights are task-specific, taking advantage of specific characteristics of the task to determine a suitable constraint on the latent space (e.g., imposing smoothness, imposing a information-prioritization scheme, or capturing a specific explanatory factor of interest, et cetera). Our demonstrations broadly cover task categories ranging across domain alignment, control, and generative modeling---thus demonstrating the ubiquitous efficacy of designing and imposing task-specific constraints during representation learning.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2022; ©2022 |
| Publication date | 2022; 2022 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Shu, Rui |
|---|---|
| Degree supervisor | Ermon, Stefano |
| Thesis advisor | Ermon, Stefano |
| Thesis advisor | Bui, Hung |
| Thesis advisor | Kochenderfer, Mykel J, 1980- |
| Thesis advisor | Ma, Tengyu |
| Degree committee member | Bui, Hung |
| Degree committee member | Kochenderfer, Mykel J, 1980- |
| Degree committee member | Ma, Tengyu |
| Associated with | Stanford University, Computer Science Department |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Rui Shu. |
|---|---|
| Note | Submitted to the Computer Science Department. |
| Thesis | Thesis Ph.D. Stanford University 2022. |
| Location | https://purl.stanford.edu/fs326sj8134 |
Access conditions
- Copyright
- © 2022 by Rui Shu
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...