Stochastic Bigger Subspace Algorithms for Nonconvex Stochastic Optimization

It is well known that the stochastic optimization problem can be regarded as one of most hard problems since, in cases, values f and its gradient are often not easily to solved, or F(∙, ξ) normally given clearly (or) distribution function P equivocal. Then an effective algorithm successfully designed used solve this interesting work. This paper designs bigger subspace algorithms for solving nonconvex problems. A general framework such presented convergence analysis, where so-called sufficient descent property, trust region feature, global stationary points proved under suitable conditions. In worst-case, we will turn out complexity competitive a accuracy parameter. We SFO-calls with diminishing steplength O(ϵ-1/1-β) random constant O(ϵ-2) respectively, β ∈ (0.5,1) ϵ needed conditions weaker than quasi-Newton methods normal conjugate algorithms. The detail variance reduction also proposed experiments binary classification done demonstrate performance algorithm.