English

Momentum-based variance-reduced proximal stochastic gradient method for composite nonconvex stochastic optimization

Optimization and Control 2022-05-02 v3 Machine Learning Numerical Analysis Numerical Analysis

Abstract

Stochastic gradient methods (SGMs) have been extensively used for solving stochastic problems or large-scale machine learning problems. Recent works employ various techniques to improve the convergence rate of SGMs for both convex and nonconvex cases. Most of them require a large number of samples in some or all iterations of the improved SGMs. In this paper, we propose a new SGM, named PStorm, for solving nonconvex nonsmooth stochastic problems. With a momentum-based variance reduction technique, PStorm can achieve the optimal complexity result O(ε3)O(\varepsilon^{-3}) to produce a stochastic ε\varepsilon-stationary solution, if a mean-squared smoothness condition holds. Different from existing optimal methods, PStorm can achieve the O(ε3){O}(\varepsilon^{-3}) result by using only one or O(1)O(1) samples in every update. With this property, PStorm can be applied to online learning problems that favor real-time decisions based on one or O(1)O(1) new observations. In addition, for large-scale machine learning problems, PStorm can generalize better by small-batch training than other optimal methods that require large-batch training and the vanilla SGM, as we demonstrate on training a sparse fully-connected neural network and a sparse convolutional neural network.

Keywords

Cite

@article{arxiv.2006.00425,
  title  = {Momentum-based variance-reduced proximal stochastic gradient method for composite nonconvex stochastic optimization},
  author = {Yangyang Xu and Yibo Xu},
  journal= {arXiv preprint arXiv:2006.00425},
  year   = {2022}
}

Comments

Yibo Xu joined this work by adding a new proof for the case with constant stepsize and removing the requirement of a large initial minibatch

R2 v1 2026-06-23T15:56:15.352Z