A Unified Convergence Theorem for Stochastic Optimization Methods

Abstract

In this work, we provide a fundamental unified convergence theorem used for deriving expected and almost sure convergence results for a series of stochastic optimization methods. Our unified theorem only requires to verify several representative conditions and is not tailored to any specific algorithm. As a direct application, we recover expected and almost sure convergence results of the stochastic gradient method (SGD) and random reshuffling (RR) under more general settings. Moreover, we establish new expected and almost sure convergence results for the stochastic proximal gradient method (prox-SGD) and stochastic model-based methods for nonsmooth nonconvex optimization problems. These applications reveal that our unified theorem provides a plugin-type convergence analysis and strong convergence guarantees for a wide class of stochastic optimization methods.

PDF NeurIPS OpenReview Semantic Scholar

Cite

Text

Li and Milzarek. "A Unified Convergence Theorem for Stochastic Optimization Methods." Neural Information Processing Systems, 2022.

Markdown

[Li and Milzarek. "A Unified Convergence Theorem for Stochastic Optimization Methods." Neural Information Processing Systems, 2022.](https://mlanthology.org/neurips/2022/li2022neurips-unified/)

BibTeX

@inproceedings{li2022neurips-unified,
  title     = {{A Unified Convergence Theorem for Stochastic Optimization Methods}},
  author    = {Li, Xiao and Milzarek, Andre},
  booktitle = {Neural Information Processing Systems},
  year      = {2022},
  url       = {https://mlanthology.org/neurips/2022/li2022neurips-unified/}
}