Smoothed Variable Sample-size Accelerated Proximal Methods for Nonsmooth Stochastic Convex Programs
Abstract
We consider minimizing when is possibly nonsmooth and either strongly convex or convex in . (I) Strongly convex. When is strongly convex in , we propose a variable sample-size accelerated proximal scheme (VS-APM) and apply it on , the (-)Moreau smoothed variant of ; we term such a scheme as (m-VS-APM). We consider three settings. (a) Bounded domains. In this setting, VS-APM displays linear convergence in inexact gradient steps, each of which requires utilizing an inner (SSG) scheme. Specifically, mVS-APM achieves an optimal oracle complexity in SSG steps; (b) Unbounded domains. In this regime, under a weaker assumption of suitable state-dependent bounds on subgradients, an unaccelerated variant mVS-PM is linearly convergent; (c) Smooth ill-conditioned . When is -smooth and , we employ mVS-APM where increasingly accurate gradients are obtained by VS-APM. Notably, mVS-APM displays linear convergence and near-optimal complexity in inner proximal evaluations (upto a log factor) compared to VS-APM. But, unlike a direct application of VS-APM, this scheme is characterized by larger steplengths and better empirical behavior; (II) Convex. When is merely convex but smoothable, by suitable choices of the smoothing, steplength, and batch-size sequences, smoothed VS-APM (or sVS-APM) produces sequences for which expected sub-optimality diminishes at the rate of with an optimal oracle complexity of . Finally, sVS-APM and VS-APM produce sequences that converge almost surely to a solution of the original problem.
Cite
@article{arxiv.1803.00718,
title = {Smoothed Variable Sample-size Accelerated Proximal Methods for Nonsmooth Stochastic Convex Programs},
author = {Afrooz Jalilzadeh and Uday V. Shanbhag and Jose H. Blanchet and Peter W. Glynn},
journal= {arXiv preprint arXiv:1803.00718},
year = {2022}
}
Comments
Stochastic Systems (2022)