Exploiting Compositionality to Explore a Large Space of Model Structures
The recent proliferation of richly structured probabilistic models raises the question of how to automatically determine an appropriate model for a dataset. We investigate this question for a space of matrix decomposition models which can express a variety of widely used models from unsupervised lea...
| Main Authors: | , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | en_US |
| Published: |
AUAI Press
2014
|
| Online Access: | http://hdl.handle.net/1721.1/86219 https://orcid.org/0000-0002-1925-2035 https://orcid.org/0000-0002-2231-7995 |
| _version_ | 1826208792800395264 |
|---|---|
| author | Grosse, Roger Baker Salakhutdinov, Ruslan Freeman, William T. Tenenbaum, Joshua B. |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Grosse, Roger Baker Salakhutdinov, Ruslan Freeman, William T. Tenenbaum, Joshua B. |
| author_sort | Grosse, Roger Baker |
| collection | MIT |
| description | The recent proliferation of richly structured probabilistic models raises the question of how to automatically determine an appropriate model for a dataset. We investigate this question for a space of matrix decomposition models which can express a variety of widely used models from unsupervised learning. To enable model selection, we organize these models into a context-free grammar which generates a wide variety of structures through the compositional application of a few simple rules. We use our grammar to generically and efficiently infer latent components and estimate predictive likelihood for nearly 2500 structures using a small toolbox of reusable algorithms. Using a greedy search over our grammar, we automatically choose the decomposition structure from raw data by evaluating only a small fraction of all models. The proposed method typically finds the correct structure for synthetic data and backs off gracefully to simpler models under heavy noise. It learns sensible structures for datasets as diverse as image patches, motion capture, 20 Questions, and U.S. Senate votes, all using exactly the same code. |
| first_indexed | 2024-09-23T14:12:43Z |
| format | Article |
| id | mit-1721.1/86219 |
| institution | Massachusetts Institute of Technology |
| language | en_US |
| last_indexed | 2024-09-23T14:12:43Z |
| publishDate | 2014 |
| publisher | AUAI Press |
| record_format | dspace |
| spelling | mit-1721.1/862192022-10-01T19:45:22Z Exploiting Compositionality to Explore a Large Space of Model Structures Grosse, Roger Baker Salakhutdinov, Ruslan Freeman, William T. Tenenbaum, Joshua B. Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Grosse, Roger Baker Freeman, William T. Tenenbaum, Joshua B. The recent proliferation of richly structured probabilistic models raises the question of how to automatically determine an appropriate model for a dataset. We investigate this question for a space of matrix decomposition models which can express a variety of widely used models from unsupervised learning. To enable model selection, we organize these models into a context-free grammar which generates a wide variety of structures through the compositional application of a few simple rules. We use our grammar to generically and efficiently infer latent components and estimate predictive likelihood for nearly 2500 structures using a small toolbox of reusable algorithms. Using a greedy search over our grammar, we automatically choose the decomposition structure from raw data by evaluating only a small fraction of all models. The proposed method typically finds the correct structure for synthetic data and backs off gracefully to simpler models under heavy noise. It learns sensible structures for datasets as diverse as image patches, motion capture, 20 Questions, and U.S. Senate votes, all using exactly the same code. United States. Army Research Office (ARO grant W911NF-08-1-0242) American Society for Engineering Education. National Defense Science and Engineering Graduate Fellowship 2014-04-23T16:23:51Z 2014-04-23T16:23:51Z 2012-08 Article http://purl.org/eprint/type/ConferencePaper 978-0-9749039-8-9 http://hdl.handle.net/1721.1/86219 Grosse, Roger B., Ruslan Salakhutdinov, William T. Freeman, and Joshua B. Tenenbaum. "Exploiting Compositionality to Explore a Large Space of Model Structures." In 28th Conference on Uncertainly in Artificial Intelligence (2012), Catalina Island, United States, August 15-17, 2012. AUAI Press, pp. 306-315. https://orcid.org/0000-0002-1925-2035 https://orcid.org/0000-0002-2231-7995 en_US http://www.auai.org/uai2012/proceedings.pdf Proceedings of the 28th Conference on Uncertainly in Artificial Intelligence (2012) Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf AUAI Press MIT web domain |
| spellingShingle | Grosse, Roger Baker Salakhutdinov, Ruslan Freeman, William T. Tenenbaum, Joshua B. Exploiting Compositionality to Explore a Large Space of Model Structures |
| title | Exploiting Compositionality to Explore a Large Space of Model Structures |
| title_full | Exploiting Compositionality to Explore a Large Space of Model Structures |
| title_fullStr | Exploiting Compositionality to Explore a Large Space of Model Structures |
| title_full_unstemmed | Exploiting Compositionality to Explore a Large Space of Model Structures |
| title_short | Exploiting Compositionality to Explore a Large Space of Model Structures |
| title_sort | exploiting compositionality to explore a large space of model structures |
| url | http://hdl.handle.net/1721.1/86219 https://orcid.org/0000-0002-1925-2035 https://orcid.org/0000-0002-2231-7995 |
| work_keys_str_mv | AT grosserogerbaker exploitingcompositionalitytoexplorealargespaceofmodelstructures AT salakhutdinovruslan exploitingcompositionalitytoexplorealargespaceofmodelstructures AT freemanwilliamt exploitingcompositionalitytoexplorealargespaceofmodelstructures AT tenenbaumjoshuab exploitingcompositionalitytoexplorealargespaceofmodelstructures |