A stochastic alternating direction method of multipliers for non-smooth and non-convex optimization

Alternating direction method of multipliers (ADMM) is a popular first-order owing to its simplicity and efficiency. However, similar other proximal splitting methods, the performance ADMM degrades significantly when scale optimization problems solve becomes large. In this paper, we consider combining with class stochastic gradient variance reduction for solving large-scale non-convex non-smooth problems. Global convergence generated sequence established under extra additional assumption that object function satisfies Kurdyka-Lojasiewicz (KL) property. Numerical experiments on graph-guided fused Lasso computed tomography are presented demonstrate proposed methods.