English

Constant Stepsize Local GD for Logistic Regression: Acceleration by Instability

Machine Learning 2025-06-18 v1

Abstract

Existing analysis of Local (Stochastic) Gradient Descent for heterogeneous objectives requires stepsizes η1/K\eta \leq 1/K where KK is the communication interval, which ensures monotonic decrease of the objective. In contrast, we analyze Local Gradient Descent for logistic regression with separable, heterogeneous data using any stepsize η>0\eta > 0. With RR communication rounds and MM clients, we show convergence at a rate O(1/ηKR)\mathcal{O}(1/\eta K R) after an initial unstable phase lasting for O~(ηKM)\widetilde{\mathcal{O}}(\eta K M) rounds. This improves upon the existing O(1/R)\mathcal{O}(1/R) rate for general smooth, convex objectives. Our analysis parallels the single machine analysis of~\cite{wu2024large} in which instability is caused by extremely large stepsizes, but in our setting another source of instability is large local updates with heterogeneous objectives.

Keywords

Cite

@article{arxiv.2506.13974,
  title  = {Constant Stepsize Local GD for Logistic Regression: Acceleration by Instability},
  author = {Michael Crawshaw and Blake Woodworth and Mingrui Liu},
  journal= {arXiv preprint arXiv:2506.13974},
  year   = {2025}
}

Comments

ICML 2025

R2 v1 2026-07-01T03:20:40.694Z