AI RESEARCH
Pessimism-Free Offline Learning in General-Sum Games via KL Regularization
arXiv CS.LG
•
ArXi:2605.00264v1 Announce Type: new Offline multi-agent reinforcement learning in general-sum settings is challenged by the distribution shift between logged datasets and target equilibrium policies. While standard methods rely on manual pessimistic penalties, we nstrate that KL regularization suffices to stabilize learning and achieve equilibrium recovery. We propose General-sum Anchored Nash Equilibrium (GANE), which recovers regularized Nash equilibria at an accelerated statistical rate of $\widetilde{O}(1/n.