XDO: A Double Oracle Algorithm for Extensive-Form Games

被引:0
|
作者
McAleer, Stephen [1 ]
Lanier, John [1 ]
Wang, Kevin A. [1 ]
Baldi, Pierre [1 ]
Fox, Roy [1 ]
机构
[1] Univ Calif Irvine, Dept Comp Sci, Irvine, CA 92717 USA
来源
ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 34 (NEURIPS 2021) | 2021年
关键词
LEVEL;
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Policy Space Response Oracles (PSRO) is a reinforcement learning (RL) algorithm for two-player zero-sum games that has been empirically shown to find approximate Nash equilibria in large games. Although PSRO is guaranteed to converge to an approximate Nash equilibrium and can handle continuous actions, it may take an exponential number of iterations as the number of information states (infostates) grows. We propose Extensive-Form Double Oracle (XDO), an extensive-form double oracle algorithm for two-player zero-sum games that is guaranteed to converge to an approximate Nash equilibrium linearly in the number of infostates. Unlike PSRO, which mixes best responses at the root of the game, XDO mixes best responses at every infostate. We also introduce Neural XDO (NXDO), where the best response is learned through deep RL. In tabular experiments on Leduc poker, we find that XDO achieves an approximate Nash equilibrium in a number of iterations an order of magnitude smaller than PSRO. Experiments on a modified Leduc poker game and Oshi-Zumo show that tabular XDO achieves a lower exploitability than CFR with the same amount of computation. We also find that NXDO outperforms PSRO and NFSP on a sequential multidimensional continuous-action game. NXDO is the first deep RL method that can find an approximate Nash equilibrium in high-dimensional continuous-action sequential games. Experiment code is available at https://github.com/indylab/nxdo.
引用
收藏
页数:12
相关论文
共 50 条
  • [1] An Exact Double-Oracle Algorithm for Zero-Sum Extensive-Form Games with Imperfect Information
    Bosansky, Branislav
    Kiekintveld, Christopher
    Lisy, Viliam
    Pechoucek, Michal
    JOURNAL OF ARTIFICIAL INTELLIGENCE RESEARCH, 2014, 51 : 829 - 866
  • [2] RATIONALITY IN EXTENSIVE-FORM GAMES
    RENY, PJ
    JOURNAL OF ECONOMIC PERSPECTIVES, 1992, 6 (04): : 103 - 118
  • [3] Timeability of Extensive-Form Games
    Jakobsen, Sune K.
    Sorensen, Troels B.
    Conitzer, Vincent
    ITCS'16: PROCEEDINGS OF THE 2016 ACM CONFERENCE ON INNOVATIONS IN THEORETICAL COMPUTER SCIENCE, 2016, : 191 - 199
  • [4] Computational Extensive-Form Games
    Halpern, Joseph Y.
    Pass, Rafael
    Seeman, Lior
    EC'16: PROCEEDINGS OF THE 2016 ACM CONFERENCE ON ECONOMICS AND COMPUTATION, 2016, : 681 - 698
  • [5] Quantum extensive-form games
    Kazuki Ikeda
    Quantum Information Processing, 22
  • [6] Quantum extensive-form games
    Ikeda, Kazuki
    QUANTUM INFORMATION PROCESSING, 2023, 22 (01)
  • [7] Strategic negotiations for extensive-form games
    Dave de Jonge
    Dongmo Zhang
    Autonomous Agents and Multi-Agent Systems, 2020, 34
  • [8] Coarse Correlation in Extensive-Form Games
    Farina, Gabriele
    Bianchi, Tommaso
    Sandholm, Tuomas
    THIRTY-FOURTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, THE THIRTY-SECOND INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE AND THE TENTH AAAI SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2020, 34 : 1934 - 1941
  • [9] Strategic negotiations for extensive-form games
    de Jonge, Dave
    Zhang, Dongmo
    AUTONOMOUS AGENTS AND MULTI-AGENT SYSTEMS, 2020, 34 (01)
  • [10] Stackelberg Equilibrium Approximation in General-Sum Extensive-Form Games with Double-Oracle Sampling Method
    Karwowski, Jan
    Mandziuk, Jacek
    AAMAS '19: PROCEEDINGS OF THE 18TH INTERNATIONAL CONFERENCE ON AUTONOMOUS AGENTS AND MULTIAGENT SYSTEMS, 2019, : 2045 - 2047