Abstract:In this paper we consider convex optimization problems with stochastic composite objective function subject to (possibly) infinite intersection of constraints. The objective function is expressed in terms of expectation operator over a sum of two terms satisfying a stochastic bounded gradient condition, with or without strong convexity type properties. In contrast to the classical approach, where the constraints are usually represented as intersection of simple sets, in this paper we consider that each constraint set is given as the level set of a convex but not necessarily differentiable function. Based on the flexibility offered by our general optimization model we consider a stochastic subgradient method with random feasibility updates. At each iteration, our algorithm takes a stochastic proximal (sub)gradient step aimed at minimizing the objective function and then a subsequent subgradient step minimizing the feasibility violation of the observed random constraint. We analyze the convergence behavior of the proposed algorithm for diminishing stepsizes and for the case when the objective function is convex or strongly convex, unifying the nonsmooth and smooth cases. We prove sublinear convergence rates for this stochastic subgradient algorithm, which are known to be optimal for subgradient methods on this class of problems. When the objective function has a linear least-square form and the constraints are polyhedral, it is shown that the algorithm converges linearly. Numerical evidence supports the effectiveness of our method in real problems.

A Smoothing Stochastic Gradient Method for Composite Optimization

Stochastic smoothing accelerated gradient method for nonsmooth convex composite optimization

Stochastic Methods for Composite Optimization Problems

Adaptive smoothing mini-batch stochastic accelerated gradient method for nonsmooth convex stochastic composite optimization

Asynchronous Proximal Stochastic Gradient Algorithm for Composition Optimization Problems

Gradient sliding for composite optimization

Stochastic subgradient for composite optimization with functional constraints

A Stochastic Nesterov’s Smoothing Accelerated Method for General Nonsmooth Constrained Stochastic Composite Convex Optimization

A Distributed Stochastic Proximal-Gradient Algorithm for Composite Optimization

On Stochastic Primal-Dual Hybrid Gradient Approach for Compositely Regularized Minimization.

Variance-Reduced Proximal Stochastic Gradient Descent for Non-convex Composite optimization.

A Variable Smoothing for Weakly Convex Composite Minimization with Nonconvex Constraint

Accelerated Method for Stochastic Composition Optimization With Nonsmooth Regularization

On the Convergence of Stochastic Compositional Gradient Descent Ascent Method.

Mini-batch stochastic Nesterov's smoothing method for constrained convex stochastic composite optimization

A Distributed Nesterov-Like Gradient Tracking Algorithm for Composite Constrained Optimization

On inexact stochastic splitting methods for a class of nonconvex composite optimization problems with relative error

On the local convergence of the semismooth Newton method for composite optimization

Adaptive Stochastic Optimisation of Nonconvex Composite Objectives

Almost sure convergence of stochastic composite objective mirror descent for non-convex non-smooth optimization

An Edge-based Stochastic Proximal Gradient Algorithm for Decentralized Composite Optimization