Title:Computing Ex Ante Coordinated Team-Maxmin Equilibria in Zero-Sum Multiplayer Extensive-Form Games

Abstract:Developing efficient algorithms to compute equilibria in multiplayer games remains an open challenge in computational game theory. In this paper, we focus on computing Team-Maxmin Equilibria with Coordination device (TMEsCor) in zero-sum multiplayer extensive-form games, where a team of players with ex ante coordination play against an this http URL can capture many real-world scenarios, including: 1) collusion in card games (e.g., Bridge and Poker), where team members work together against a target player but cannot communicate during the game; and 2) coordination in protecting forests, where coordinated groups cannot communicate during interdicting illegal loggers. Existing algorithms are inefficient to compute TMEsCor in large games, e.g., they can compute exact TMEsCor only for very small three-player Kuhn poker games. To compute TMEsCor efficiently, we first propose anew hybrid-form strategy representation for the team, which does not change the set of equilibria in a game. Second, we develop a column generation algorithm converging within a finite number of iterations in the infinite strategy space and containing a novel best response algorithm with multilinear terms representing reaching probabilities for terminal nodes to reduce the number of variables. Third, we develop a novel associated representation technique exactly representing multilinear terms by linear constraints and efficiently generating associated constraints for the equivalence relation between multilinear terms in the game tree. Finally, experimental results show that our algorithm is several orders of magnitude faster than prior state-of-the-art algorithms in large games.