Combining Deep Reinforcement Learning and Search for Imperfect-Information Games

https://arxiv.org/abs/2007.13544

ReBeL: self-play RL plus search that provably converges to a Nash equilibrium in two-player zero-sum games; reduces to an AlphaZero-like algorithm in perfect-information games; superhuman heads-up no-limit hold'em.

Regret minimization and search for imperfect-information games