English

Perturbed Iterate SGD for Lipschitz Continuous Loss Functions

Optimization and Control 2022-10-05 v5

Abstract

This paper presents an extension of stochastic gradient descent for the minimization of Lipschitz continuous loss functions. Our motivation is for use in non-smooth non-convex stochastic optimization problems, which are frequently encountered in applications such as machine learning. Using the Clarke ϵ\epsilon-subdifferential, we prove the non-asymptotic convergence to an approximate stationary point in expectation for the proposed method. From this result, a method with non-asymptotic convergence with high probability, as well as a method with asymptotic convergence to a Clarke stationary point almost surely are developed. Our results hold under the assumption that the stochastic loss function is a Carath\'eodory function which is almost everywhere Lipschitz continuous in the decision variables. To the best of our knowledge this is the first non-asymptotic convergence analysis under these minimal assumptions.

Keywords

Cite

@article{arxiv.2003.07606,
  title  = {Perturbed Iterate SGD for Lipschitz Continuous Loss Functions},
  author = {Michael R. Metel and Akiko Takeda},
  journal= {arXiv preprint arXiv:2003.07606},
  year   = {2022}
}
R2 v1 2026-06-23T14:17:09.046Z