All learning is local: Multi-agent learning in global reward games
In large multiagent games, partial observability, coordination, and credit assignment persistently plague attempts to design good learning algorithms. We provide a simple and efficient algorithm that in part uses a linear system to model the world from a single agent’s limited perspective, and takes...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | en_US |
| Published: |
2003
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/1721.1/3851 |
| _version_ | 1826209006570438656 |
|---|---|
| author | Chang, Yu-Han Ho, Tracey Kaelbling, Leslie P. |
| author_facet | Chang, Yu-Han Ho, Tracey Kaelbling, Leslie P. |
| author_sort | Chang, Yu-Han |
| collection | MIT |
| description | In large multiagent games, partial observability, coordination, and credit assignment persistently plague attempts to design good learning algorithms. We provide a simple and efficient algorithm that in part uses a linear system to model the world from a single agent’s limited perspective, and takes advantage of Kalman filtering to allow an agent to construct a good training signal and effectively learn a near-optimal policy in a wide variety of settings. A sequence of increasingly complex empirical tests verifies the efficacy of this technique. |
| first_indexed | 2024-09-23T14:15:52Z |
| format | Article |
| id | mit-1721.1/3851 |
| institution | Massachusetts Institute of Technology |
| language | en_US |
| last_indexed | 2024-09-23T14:15:52Z |
| publishDate | 2003 |
| record_format | dspace |
| spelling | mit-1721.1/38512019-04-12T11:15:05Z All learning is local: Multi-agent learning in global reward games Chang, Yu-Han Ho, Tracey Kaelbling, Leslie P. Kalman filtering multi-agent systems Q-learning reinforcement learning In large multiagent games, partial observability, coordination, and credit assignment persistently plague attempts to design good learning algorithms. We provide a simple and efficient algorithm that in part uses a linear system to model the world from a single agent’s limited perspective, and takes advantage of Kalman filtering to allow an agent to construct a good training signal and effectively learn a near-optimal policy in a wide variety of settings. A sequence of increasingly complex empirical tests verifies the efficacy of this technique. Singapore-MIT Alliance (SMA) 2003-12-13T18:55:17Z 2003-12-13T18:55:17Z 2004-01 Article http://hdl.handle.net/1721.1/3851 en_US Computer Science (CS); 1408858 bytes application/pdf application/pdf |
| spellingShingle | Kalman filtering multi-agent systems Q-learning reinforcement learning Chang, Yu-Han Ho, Tracey Kaelbling, Leslie P. All learning is local: Multi-agent learning in global reward games |
| title | All learning is local: Multi-agent learning in global reward games |
| title_full | All learning is local: Multi-agent learning in global reward games |
| title_fullStr | All learning is local: Multi-agent learning in global reward games |
| title_full_unstemmed | All learning is local: Multi-agent learning in global reward games |
| title_short | All learning is local: Multi-agent learning in global reward games |
| title_sort | all learning is local multi agent learning in global reward games |
| topic | Kalman filtering multi-agent systems Q-learning reinforcement learning |
| url | http://hdl.handle.net/1721.1/3851 |
| work_keys_str_mv | AT changyuhan alllearningislocalmultiagentlearninginglobalrewardgames AT hotracey alllearningislocalmultiagentlearninginglobalrewardgames AT kaelblinglesliep alllearningislocalmultiagentlearninginglobalrewardgames |