Learning Tree Structured Potential Games
© 2016 NIPS Foundation - All Rights Reserved. Many real phenomena, including behaviors, involve strategic interactions that can be learned from data. We focus on learning tree structured potential games where equilibria are represented by local maxima of an underlying potential function. We cast the...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137609 |
| _version_ | 1826213387087904768 |
|---|---|
| author | Garg, Vikas K. Jaakkola, Tommi |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Garg, Vikas K. Jaakkola, Tommi |
| author_sort | Garg, Vikas K. |
| collection | MIT |
| description | © 2016 NIPS Foundation - All Rights Reserved. Many real phenomena, including behaviors, involve strategic interactions that can be learned from data. We focus on learning tree structured potential games where equilibria are represented by local maxima of an underlying potential function. We cast the learning problem within a max margin setting and show that the problem is NP-hard even when the strategic interactions form a tree. We develop a variant of dual decomposition to estimate the underlying game and demonstrate with synthetic and real decision/voting data that the game theoretic perspective (carving out local maxima) enables meaningful recovery. |
| first_indexed | 2024-09-23T15:48:17Z |
| format | Article |
| id | mit-1721.1/137609 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T15:48:17Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | mit-1721.1/1376092021-11-06T03:04:45Z Learning Tree Structured Potential Games Garg, Vikas K. Jaakkola, Tommi Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory © 2016 NIPS Foundation - All Rights Reserved. Many real phenomena, including behaviors, involve strategic interactions that can be learned from data. We focus on learning tree structured potential games where equilibria are represented by local maxima of an underlying potential function. We cast the learning problem within a max margin setting and show that the problem is NP-hard even when the strategic interactions form a tree. We develop a variant of dual decomposition to estimate the underlying game and demonstrate with synthetic and real decision/voting data that the game theoretic perspective (carving out local maxima) enables meaningful recovery. 2021-11-05T19:57:32Z 2021-11-05T19:57:32Z 2016 2019-05-31T16:14:13Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137609 Garg, Vikas K. and Jaakkola, Tommi. 2016. "Learning Tree Structured Potential Games." en Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Neural Information Processing Systems (NIPS) |
| spellingShingle | Garg, Vikas K. Jaakkola, Tommi Learning Tree Structured Potential Games |
| title | Learning Tree Structured Potential Games |
| title_full | Learning Tree Structured Potential Games |
| title_fullStr | Learning Tree Structured Potential Games |
| title_full_unstemmed | Learning Tree Structured Potential Games |
| title_short | Learning Tree Structured Potential Games |
| title_sort | learning tree structured potential games |
| url | https://hdl.handle.net/1721.1/137609 |
| work_keys_str_mv | AT gargvikask learningtreestructuredpotentialgames AT jaakkolatommi learningtreestructuredpotentialgames |