Learning hierarchical teaching policies for cooperative agents
© 2020 International Foundation for Autonomous. Collective learning can be greatly enhanced when agents effectively exchange knowledge with their peers. In particular, recent work studying agents that learn to teach other teammates has demonstrated that action advising accelerates team-wide learning...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137164 |
| _version_ | 1826217997084131328 |
|---|---|
| author | How, Jonathan P. |
| author2 | MIT-IBM Watson AI Lab |
| author_facet | MIT-IBM Watson AI Lab How, Jonathan P. |
| author_sort | How, Jonathan P. |
| collection | MIT |
| description | © 2020 International Foundation for Autonomous. Collective learning can be greatly enhanced when agents effectively exchange knowledge with their peers. In particular, recent work studying agents that learn to teach other teammates has demonstrated that action advising accelerates team-wide learning. However, the prior work has simplified the learning of advising policies by using simple function approximations and only considered advising with primitive (low-level) actions, limiting the scalability of learning and teaching to complex domains. This paper introduces a novel learning-to-teach framework, called hierarchical multiagent teaching (HMAT), that improves scalability to complex environments by using the deep representation for student policies and by advising with more expressive extended action sequences over multiple levels of temporal abstraction. Our empirical evaluations demonstrate that HMAT improves team-wide learning progress in large, complex domains where previous approaches fail. HMAT also learns teaching policies that can effectively transfer knowledge to different teammates with knowledge of different tasks, even when the teammates have heterogeneous action spaces. |
| first_indexed | 2024-09-23T17:12:30Z |
| format | Article |
| id | mit-1721.1/137164 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T17:12:30Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | mit-1721.1/1371642023-02-09T16:21:15Z Learning hierarchical teaching policies for cooperative agents How, Jonathan P. MIT-IBM Watson AI Lab Massachusetts Institute of Technology. Laboratory for Information and Decision Systems © 2020 International Foundation for Autonomous. Collective learning can be greatly enhanced when agents effectively exchange knowledge with their peers. In particular, recent work studying agents that learn to teach other teammates has demonstrated that action advising accelerates team-wide learning. However, the prior work has simplified the learning of advising policies by using simple function approximations and only considered advising with primitive (low-level) actions, limiting the scalability of learning and teaching to complex domains. This paper introduces a novel learning-to-teach framework, called hierarchical multiagent teaching (HMAT), that improves scalability to complex environments by using the deep representation for student policies and by advising with more expressive extended action sequences over multiple levels of temporal abstraction. Our empirical evaluations demonstrate that HMAT improves team-wide learning progress in large, complex domains where previous approaches fail. HMAT also learns teaching policies that can effectively transfer knowledge to different teammates with knowledge of different tasks, even when the teammates have heterogeneous action spaces. 2021-11-02T18:45:14Z 2021-11-02T18:45:14Z 2020 2021-04-30T14:19:30Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137164 How, Jonathan P. 2020. "Learning hierarchical teaching policies for cooperative agents." Proceedings of the International Joint Conference on Autonomous Agents and Multiagent Systems, AAMAS, 2020-May. en https://dl.acm.org/doi/10.5555/3398761.3398836 Proceedings of the International Joint Conference on Autonomous Agents and Multiagent Systems, AAMAS Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf arXiv |
| spellingShingle | How, Jonathan P. Learning hierarchical teaching policies for cooperative agents |
| title | Learning hierarchical teaching policies for cooperative agents |
| title_full | Learning hierarchical teaching policies for cooperative agents |
| title_fullStr | Learning hierarchical teaching policies for cooperative agents |
| title_full_unstemmed | Learning hierarchical teaching policies for cooperative agents |
| title_short | Learning hierarchical teaching policies for cooperative agents |
| title_sort | learning hierarchical teaching policies for cooperative agents |
| url | https://hdl.handle.net/1721.1/137164 |
| work_keys_str_mv | AT howjonathanp learninghierarchicalteachingpoliciesforcooperativeagents |