Online Minimax Q Network Learning for Two-Player Zero-Sum Markov Games

doi:10.1109/TNNLS.2020.3041469

CORC > 自动化研究所 > 中国科学院自动化研究所 > 复杂系统管理与控制国家重点实验室 > 深度强化学习

	Online Minimax Q Network Learning for Two-Player Zero-Sum Markov Games
	Zhu, Yuanheng 1,2; Zhao, Dongbin 1,2
刊名	IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
	2022-03-01
卷号	33 期号:3 页码:1228-1241
关键词	Games Nash equilibrium Mathematical model Markov processes Convergence Dynamic programming Training Deep reinforcement learning (DRL) generalized policy iteration (GPI) Markov game (MG) Nash equilibrium Q network zero sum
ISSN号	2162-237X
DOI	10.1109/TNNLS.2020.3041469
通讯作者	Zhao, Dongbin(dongbin.zhao@ia.ac.cn)
英文摘要	The Nash equilibrium is an important concept in game theory. It describes the least exploitability of one player from any opponents. We combine game theory, dynamic programming, and recent deep reinforcement learning (DRL) techniques to online learn the Nash equilibrium policy for two-player zero-sum Markov games (TZMGs). The problem is first formulated as a Bellman minimax equation, and generalized policy iteration (GPI) provides a double-loop iterative way to find the equilibrium. Then, neural networks are introduced to approximate Q functions for large-scale problems. An online minimax Q network learning algorithm is proposed to train the network with observations. Experience replay, dueling network, and double Q-learning are applied to improve the learning process. The contributions are twofold: 1) DRL techniques are combined with GPI to find the TZMG Nash equilibrium for the first time and 2) the convergence of the online learning algorithm with a lookup table and experience replay is proven, whose proof is not only useful for TZMGs but also instructive for single-agent Markov decision problems. Experiments on different examples validate the effectiveness of the proposed algorithm on TZMG problems.
资助项目	National Key Research and Development Program of China[2018AAA0102404] ; National Key Research and Development Program of China[2018AAA0101005]
WOS关键词	NONLINEAR-SYSTEMS ; GO ; ALGORITHM ; LEVEL
WOS研究方向	Computer Science ; Engineering
语种	英语
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
WOS记录号	WOS:000766269100030
资助机构	National Key Research and Development Program of China
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/48234]
专题	复杂系统管理与控制国家重点实验室_深度强化学习
通讯作者	Zhao, Dongbin
作者单位	1.Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing 100049, Peoples R China 2.Chinese Acad Sci, Inst Automat, State Key Lab Management & Control Complex Syst, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Zhu, Yuanheng,Zhao, Dongbin. Online Minimax Q Network Learning for Two-Player Zero-Sum Markov Games[J]. IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS,2022,33(3):1228-1241.
APA	Zhu, Yuanheng,&Zhao, Dongbin.(2022).Online Minimax Q Network Learning for Two-Player Zero-Sum Markov Games.IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS,33(3),1228-1241.
MLA	Zhu, Yuanheng,et al."Online Minimax Q Network Learning for Two-Player Zero-Sum Markov Games".IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS 33.3(2022):1228-1241.