PMIC: Improving Multi-Agent Reinforcement Learning with Progressive Mutual Information Collaboration

Pengyi Li, Hongyao Tang, Tianpei Yang, Xiaotian Hao, Tong Sang, Yan Zheng, Jianye Hao, Matthew E. Taylor, Wenyuan Tao, Zhen Wang
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:12979-12997, 2022.

Abstract

Learning to collaborate is critical in Multi-Agent Reinforcement Learning (MARL). Previous works promote collaboration by maximizing the correlation of agents’ behaviors, which is typically characterized by Mutual Information (MI) in different forms. However, we reveal sub-optimal collaborative behaviors also emerge with strong correlations, and simply maximizing the MI can, surprisingly, hinder the learning towards better collaboration. To address this issue, we propose a novel MARL framework, called Progressive Mutual Information Collaboration (PMIC), for more effective MI-driven collaboration. PMIC uses a new collaboration criterion measured by the MI between global states and joint actions. Based on this criterion, the key idea of PMIC is maximizing the MI associated with superior collaborative behaviors and minimizing the MI associated with inferior ones. The two MI objectives play complementary roles by facilitating better collaborations while avoiding falling into sub-optimal ones. Experiments on a wide range of MARL benchmarks show the superior performance of PMIC compared with other algorithms.

Cite this Paper


BibTeX
@InProceedings{pmlr-v162-li22s, title = {{PMIC}: Improving Multi-Agent Reinforcement Learning with Progressive Mutual Information Collaboration}, author = {Li, Pengyi and Tang, Hongyao and Yang, Tianpei and Hao, Xiaotian and Sang, Tong and Zheng, Yan and Hao, Jianye and Taylor, Matthew E. and Tao, Wenyuan and Wang, Zhen}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {12979--12997}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/li22s/li22s.pdf}, url = {https://proceedings.mlr.press/v162/li22s.html}, abstract = {Learning to collaborate is critical in Multi-Agent Reinforcement Learning (MARL). Previous works promote collaboration by maximizing the correlation of agents’ behaviors, which is typically characterized by Mutual Information (MI) in different forms. However, we reveal sub-optimal collaborative behaviors also emerge with strong correlations, and simply maximizing the MI can, surprisingly, hinder the learning towards better collaboration. To address this issue, we propose a novel MARL framework, called Progressive Mutual Information Collaboration (PMIC), for more effective MI-driven collaboration. PMIC uses a new collaboration criterion measured by the MI between global states and joint actions. Based on this criterion, the key idea of PMIC is maximizing the MI associated with superior collaborative behaviors and minimizing the MI associated with inferior ones. The two MI objectives play complementary roles by facilitating better collaborations while avoiding falling into sub-optimal ones. Experiments on a wide range of MARL benchmarks show the superior performance of PMIC compared with other algorithms.} }
Endnote
%0 Conference Paper %T PMIC: Improving Multi-Agent Reinforcement Learning with Progressive Mutual Information Collaboration %A Pengyi Li %A Hongyao Tang %A Tianpei Yang %A Xiaotian Hao %A Tong Sang %A Yan Zheng %A Jianye Hao %A Matthew E. Taylor %A Wenyuan Tao %A Zhen Wang %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-li22s %I PMLR %P 12979--12997 %U https://proceedings.mlr.press/v162/li22s.html %V 162 %X Learning to collaborate is critical in Multi-Agent Reinforcement Learning (MARL). Previous works promote collaboration by maximizing the correlation of agents’ behaviors, which is typically characterized by Mutual Information (MI) in different forms. However, we reveal sub-optimal collaborative behaviors also emerge with strong correlations, and simply maximizing the MI can, surprisingly, hinder the learning towards better collaboration. To address this issue, we propose a novel MARL framework, called Progressive Mutual Information Collaboration (PMIC), for more effective MI-driven collaboration. PMIC uses a new collaboration criterion measured by the MI between global states and joint actions. Based on this criterion, the key idea of PMIC is maximizing the MI associated with superior collaborative behaviors and minimizing the MI associated with inferior ones. The two MI objectives play complementary roles by facilitating better collaborations while avoiding falling into sub-optimal ones. Experiments on a wide range of MARL benchmarks show the superior performance of PMIC compared with other algorithms.
APA
Li, P., Tang, H., Yang, T., Hao, X., Sang, T., Zheng, Y., Hao, J., Taylor, M.E., Tao, W. & Wang, Z.. (2022). PMIC: Improving Multi-Agent Reinforcement Learning with Progressive Mutual Information Collaboration. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:12979-12997 Available from https://proceedings.mlr.press/v162/li22s.html.

Related Material