Convergence and concentration properties of constant step-size SGD through Markov chains
Electronic Journal of Statistics · 01.01.2025 · 19 (2) · 5843-5894
DOI: 10.1214/25-ejs2471 ↗Kurz gefasst
Fachartikel in der Zeitschrift Electronic Journal of Statistics, erschienen am 01.01.2025. Verfasst von Ibrahim Merad und 1 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 98 Literaturangaben.
Abstract (Originalsprache)
We consider the optimization of a smooth and strongly convex objective using constant step-size stochastic gradient descent (SGD) and study its properties through the prism of Markov chains. We show that, for unbiased gradient estimates with mildly controlled variance, the iteration converges to an invariant distribution in total variation distance. We also establish this convergence in Wasserstein-2 distance under a relaxed assumption on the gradient noise distribution compared to previous work. Our analysis shows that the SGD iterates and their invariant limit distribution inherit sub-Gaussian or sub-exponential concentration properties when these hold true for the gradient. This allows the derivation of high-confidence bounds for the final estimate. Finally, under such conditions in the linear case, we obtain a dimension-free deviation bound for the Polyak-Ruppert average of a tail sequence. All our results are non-asymptotic and their consequences are discussed through a few applications.
Zitieren
APA
Merad, I., & Gaïffas, S. (2025). Convergence and concentration properties of constant step-size SGD through Markov chains. Electronic Journal of Statistics, 19(2), 5843-5894. https://doi.org/10.1214/25-ejs2471
Vancouver
Merad I, Gaïffas S. Convergence and concentration properties of constant step-size SGD through Markov chains. Electronic Journal of Statistics. 2025;19(2):5843-5894. doi:10.1214/25-ejs2471
BibTeX
@article{Merad2025,
author = {Merad, Ibrahim and Gaïffas, Stéphane},
title = {Convergence and concentration properties of constant step-size SGD through Markov chains},
journal = {Electronic Journal of Statistics},
year = {2025},
volume = {19},
number = {2},
pages = {5843--5894},
doi = {10.1214/25-ejs2471}
}
EUVIMED