arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2202.09422cs.MAcs.AIcs.GTcs.LG
收藏

齐次马尔可夫对策的通信有效Actor-Critic方法

Communication-Efficient Actor-Critic Methods for Homogeneous Markov Games

Artificial Intelligence Institute, University of South Carolina(南卡罗来纳大学人工智能研究所)

Dingyang Chen, Yile Li, Qi Zhang

摘要:

协作式多智能体强化学习(MARL)最近的成功依赖于集中式训练和策略共享。集中式训练消除了非平稳性MARL的问题,但引起了巨大的通信成本,政策共享是经验上的关键,在某些任务的有效学习,但缺乏理论依据。在本文中,我们正式刻画了合作马尔可夫博弈的一个子类,其中代理表现出一定形式的同质性,使得策略共享可证明不会产生次优性。这使我们能够开发出第一个基于共识的去中心化参与者-批评者方法,其中共识更新适用于参与者和批评者,同时确保收敛。我们还开发了基于分散式演员-评论家方法的实用算法,以降低训练过程中的通信成本,同时仍然产生与集中式训练相当的策略。

英文摘要:

Recent success in cooperative multi-agent reinforcement learning (MARL) relies on centralized training and policy sharing. Centralized training eliminates the issue of non-stationarity MARL yet induces large communication costs, and policy sharing is empirically crucial to efficient learning in certain tasks yet lacks theoretical justification. In this paper, we formally characterize a subclass of cooperative Markov games where agents exhibit a certain form of homogeneity such that policy sharing provably incurs no suboptimality. This enables us to develop the first consensus-based decentralized actor-critic method where the consensus update is applied to both the actors and the critics while ensuring convergence. We also develop practical algorithms based on our decentralized actor-critic method to reduce the communication cost during training, while still yielding policies comparable with centralized training.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑