Zeroth-Order Proximal Stochastic Recursive Momentum Algorithm for Nonconvex Nonsmooth Optimization

Abstract

In this paper, we propose a zeroth-order proximal stochastic recursive momentum algorithm(ZO-ProxSTORM) with mini-batch for solving problems where the explicit gradients of the objective function are difficult to calculate or only the objective function values are available. Then, we analyze the convergence of the proposed algorithm under suitable assumptions. Finally, some numerical experiments are given to illustrate the efficiency of the proposed algorithm.