Block-Coordinate Methods and Restarting for Solving Extensive-Form Games

被引:0
|
作者
Chakrabarti, Darshan [1 ]
Diakonikolas, Jelena [2 ]
Kroer, Christian [1 ]
机构
[1] Columbia Univ, IEOR Dept, New York, NY 10025 USA
[2] Univ Wisconsin Madison, Dept Comp Sci, Madison, WI 53706 USA
基金
美国国家科学基金会;
关键词
VARIATIONAL-INEQUALITIES; CONVERGENCE;
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
Coordinate descent methods are popular in machine learning and optimization for their simple sparse updates and excellent practical performance. In the context of large-scale sequential game solving, these same properties would be attractive, but until now no such methods were known, because the strategy spaces do not satisfy the typical separable block structure exploited by such methods. We present the first cyclic coordinate-descent-like method for the polytope of sequence-form strategies, which form the strategy spaces for the players in an extensive-form game (EFG). Our method exploits the recursive structure of the proximal update induced by what are known as dilated regularizers, in order to allow for a pseudo block-wise update. We show that our method enjoys a O(1/T) convergence rate to a two-player zero-sum Nash equilibrium, while avoiding the worst-case polynomial scaling with the number of blocks common to cyclic methods. We empirically show that our algorithm usually performs better than other state-of-the-art first-order methods (i.e., mirror prox), and occasionally can even beat CFR+, a state-of-the-art algorithm for numerical equilibrium computation in zero-sum EFGs. We then introduce a restarting heuristic for EFG solving. We show empirically that restarting can lead to speedups, sometimes huge, both for our cyclic method, as well as for existing methods such as mirror prox and predictive CFR+.
引用
收藏
页数:29
相关论文
共 50 条
  • [31] Regret-Based Pruning in Extensive-Form Games
    Brown, Noam
    Sandholm, Tuomas
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 28 (NIPS 2015), 2015, 28
  • [32] Theoretical and Practical Advances on Smoothing for Extensive-Form Games
    Kroer, Christian
    Waugh, Kevin
    Kilinc-Karzan, Fatma
    Sandholm, Tuomas
    EC'17: PROCEEDINGS OF THE 2017 ACM CONFERENCE ON ECONOMICS AND COMPUTATION, 2017, : 693 - 693
  • [33] Safe Search for Stackelberg Equilibria in Extensive-Form Games
    Ling, Chun Kai
    Brown, Noam
    THIRTY-FIFTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, THIRTY-THIRD CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE AND THE ELEVENTH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2021, 35 : 5541 - 5548
  • [34] CFR-MIX: Solving Imperfect Information Extensive-Form Games with Combinatorial Action Space
    Li, Shuxin
    Zhang, Youzhi
    Wang, Xinrun
    Xue, Wanqi
    An, Bo
    PROCEEDINGS OF THE THIRTIETH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, IJCAI 2021, 2021, : 3663 - 3669
  • [35] Incremental Strategy Generation for Stackelberg Equilibria in Extensive-Form Games
    Cerny, Jakub
    Bosansky, Branislav
    Kiekintveld, Christopher
    ACM EC'18: PROCEEDINGS OF THE 2018 ACM CONFERENCE ON ECONOMICS AND COMPUTATION, 2018, : 151 - 168
  • [36] Near-Optimal Φ-Regret Learning in Extensive-Form Games
    Anagnostides, Ioannis
    Farina, Gabriele
    Sandholm, Tuomas
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 202, 2023, 202 : 814 - 839
  • [37] Convergence of best-response dynamics in extensive-form games
    Xu, Zibo
    JOURNAL OF ECONOMIC THEORY, 2016, 162 : 21 - 54
  • [38] Faster Optimistic Online Mirror Descent for Extensive-Form Games
    Jiang, Huacong
    Liu, Weiming
    Li, Bin
    PRICAI 2022: TRENDS IN ARTIFICIAL INTELLIGENCE, PT I, 2022, 13629 : 90 - 103
  • [39] Sequence-Form Algorithm for Computing Stackelberg Equilibria in Extensive-Form Games
    Bosansky, Branislav
    Cermak, Jiri
    PROCEEDINGS OF THE TWENTY-NINTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, 2015, : 805 - 811
  • [40] Trembling-Hand Perfection in Extensive-Form Games with Commitment
    Farina, Gabriele
    Marchesi, Alberto
    Kroer, Christian
    Gatti, Nicola
    Sandholm, Tuomas
    PROCEEDINGS OF THE TWENTY-SEVENTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, 2018, : 233 - 239