Multi-Agent Training beyond Zero-Sum with Correlated Equilibrium Meta-Solvers

arXiv:2106.09435v3 Announce Type: replace-cross
Abstract: Two-player, constant-sum games are well studied in the literature, but there has been limited progress outside of this setting. We propose Joint Policy-Space Response Oracles (JPSRO), an algorithm for training agents in n-player, general-sum extensive form games, which provably converges to an equilibrium. We further suggest correlated equilibria (CE) as promising meta-solvers, and propose a novel solution concept Maximum Gini Correlated Equilibrium (MGCE), a principled and computationally efficient family of solutions for solving the correlated …

abstract agent agents algorithm arxiv beyond cs.ai cs.gt cs.lg cs.ma equilibrium form games general literature meta multi-agent policy progress space sum training type