STOCHASTIC PROXIMAL DIFFERENCE-OF-CONVEX ALGORITHM WITH SPIDER FOR A CLASS OF NONCONVEX NONSMOOTH REGULARIZED PROBLEMS

In this paper, we propose a stochastic proximal difference-of-convex algorithm with SPIDER for solving a class of the nonconvex nonsmooth regularized problems, which can be written as the sum of a smooth function, a convex nonsmooth function, and a nonconvex nonsmooth function. At each iteration, one just separately computes the proximal operators of the nonsmooth functions, rather than that of the sum of these nonsmooth functions. A notable advantage is that our method has better gradient complexity than that of variance reduction stochastic proximal algorithm. Preliminary numerical experiments show the efficiency of the proposed method.