Title:Provable Particle-based Primal-Dual Algorithm for Mixed Nash Equilibrium

Abstract:We consider the general nonconvex nonconcave minimax problem over continuous variables. A major challenge for this problem is that a saddle point may not exist. In order to resolve this difficulty, we consider the related problem of finding a Mixed Nash Equilibrium, which is a randomized strategy represented by probability distributions over the continuous variables. We propose a Particle-based Primal-Dual Algorithm (PPDA) for a weakly entropy-regularized min-max optimization procedure over the probability distributions, which employs the stochastic movements of particles to represent the updates of random strategies for the mixed Nash Equilibrium. A rigorous convergence analysis of the proposed algorithm is provided. Compared to previous works that try to update particle weights without movements, PPDA is the first implementable particle-based algorithm with non-asymptotic quantitative convergence results, running time, and sample complexity guarantees. Our framework gives new insights into the design of particle-based algorithms for continuous min-max optimization in the general nonconvex nonconcave setting.