Operationalizing language for machine learning : supervision, interpretation, and communication
Abstract/Contents
- Abstract
- Language is a uniquely powerful tool for conceptualizing the world that allows humans to teach, understand, and collaborate with each other. This dissertation surveys three ways in which language can similarly improve the performance and usability of machine learning systems across a variety of tasks and modalities. First, supervision: we use language abstractions to help regularize models and agents for improved performance, even for downstream tasks in vision and reinforcement learning that do not necessarily require language understanding. Second, interpretation: we use language and compositionality to generate explanations of the neurons inside deep neural networks, allowing us to understand and even control model behavior. Finally, communication: we study a class of multi-agent signaling games, helping agents learn more robust and interpretable languages by encouraging them to express the generalizations encoded in human languages.
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 | 2023; ©2023 |
Publication date | 2023; 2023 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Mu, Jesse L |
---|---|
Degree supervisor | Goodman, Noah (Noah D.) |
Thesis advisor | Goodman, Noah (Noah D.) |
Thesis advisor | Andreas, Jacob |
Thesis advisor | Liang, Percy |
Degree committee member | Andreas, Jacob |
Degree committee member | Liang, Percy |
Associated with | Stanford University, School of Engineering |
Associated with | Stanford University, Computer Science Department |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Jesse L. Mu. |
---|---|
Note | Submitted to the Computer Science Department. |
Thesis | Thesis Ph.D. Stanford University 2023. |
Location | https://purl.stanford.edu/tb801sk9437 |
Access conditions
- Copyright
- © 2023 by Jesse L Mu
- 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...