Equilibrium Computation Techniques for Bridge

Zero-sum extensive-form games have been widely studied in the scientific literature. For instance, the Libratus bot defeated the top 4 human players of Texas Hold’em Poker on January 2017. However, these techniques do not directly apply to games with 3 or more players, which represent nowadays one of the most challenging problems in artificial intelligence. In this work, we will focus on the Bridge game which presents 2 teams, each composed of 2 players. We will consider parametric versions of Bridge, where the parameter is the number of cards. We will study the limits of counter-factual regret minimization algorithms when applied to these games and we will develop alternative techniques to compute an equilibrium. We will focus on testing ways to estimate subgame values efficiently and effectively.