Combining Deep Reinforcement Learning and Search for Imperfect-Information Games
This addresses the problem of developing AI for complex, real-world imperfect-information games like poker, representing a significant advance beyond prior methods limited to perfect-information settings.
The paper tackles the challenge of applying deep reinforcement learning and search to imperfect-information games, presenting ReBeL, a framework that provably converges to a Nash equilibrium in two-player zero-sum games and achieves superhuman performance in heads-up no-limit Texas hold'em poker with less domain knowledge than prior AIs.
The combination of deep reinforcement learning and search at both training and test time is a powerful paradigm that has led to a number of successes in single-agent settings and perfect-information games, best exemplified by AlphaZero. However, prior algorithms of this form cannot cope with imperfect-information games. This paper presents ReBeL, a general framework for self-play reinforcement learning and search that provably converges to a Nash equilibrium in any two-player zero-sum game. In the simpler setting of perfect-information games, ReBeL reduces to an algorithm similar to AlphaZero. Results in two different imperfect-information games show ReBeL converges to an approximate Nash equilibrium. We also show ReBeL achieves superhuman performance in heads-up no-limit Texas hold'em poker, while using far less domain knowledge than any prior poker AI.