Title:Riemannian Hamiltonian methods for min-max optimization on manifolds

Abstract:In this paper, we study the min-max optimization problems on Riemannian manifolds. We introduce a Riemannian Hamiltonian function, minimization of which serves as a proxy for solving the original min-max problems. Under the Riemannian Polyak--Łojasiewicz (PL) condition on the Hamiltonian function, its minimizer corresponds to the desired min-max saddle point. We also provide cases where this condition is satisfied. To minimize the Hamiltonian function, we propose Riemannian Hamiltonian methods (RHM) and present their convergence analysis. We extend RHM to include a consensus regularization and to the stochastic setting. We illustrate the efficacy of the proposed RHM in applications such as subspace robust Wasserstein distance, robust training of neural networks, and generative adversarial networks.