Open problem: Convergence of single-timescale mean-field Langevin descent-ascent for two-player zero-sum games.

Guillaume Wang Lénaïc Chizat

Published in: COLT (2024)

Keyphrases

perfect information
optimal strategy
nash equilibrium
reinforcement learning algorithms
imperfect information
statistical mechanics
computational complexity
probability distribution
em algorithm
closed form
iterative algorithms
global convergence
free energy