Computing Perfect Bayesian Equilibria, with Application to Empirical Game-Theoretic Analysis
New algorithm adapts Counterfactual Regret Minimization to solve complex imperfect-information games with belief consistency.
Researchers Christine Konicki, Mithun Chakraborty, and Michael P. Wellman developed PBE-CFR, a scalable algorithm for computing Perfect Bayesian Equilibria (PBE) in two-player extensive-form games. It adapts Counterfactual Regret Minimization (CFR) to maintain belief consistency while minimizing regret. The method proves correct for zero-sum games and shows competent performance on medium-to-large non-zero-sum games. It enables higher-quality strategy exploration in empirical game-theoretic analysis compared to standard Nash equilibrium solvers.
Why It Matters
Enables more realistic AI for complex strategic interactions like poker, negotiations, and security games where players have hidden information.