Saved in:
Bibliographic Details
Main Authors: Cheng, Difei, Jin, Ruinan, Qiao, Hong, Zhang, Bo
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.10889
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915289789104128
author Cheng, Difei
Jin, Ruinan
Qiao, Hong
Zhang, Bo
author_facet Cheng, Difei
Jin, Ruinan
Qiao, Hong
Zhang, Bo
contents Distributed stochastic gradient methods are widely used to preserve data privacy and ensure scalability in large-scale learning tasks. While existing theory on distributed momentum Stochastic Gradient Descent (mSGD) mainly focuses on time-averaged convergence, the more practical last-iterate convergence remains underexplored. In this work, we analyze the last-iterate convergence behavior of distributed mSGD in non-convex settings under the classical Robbins-Monro step-size schedule. We prove both almost sure convergence and $L_2$ convergence of the last iterate, and derive convergence rates. We further show that momentum can accelerate early-stage convergence, and provide experiments to support our theory.
format Preprint
id arxiv_https___arxiv_org_abs_2505_10889
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Convergence Analysis of the Last Iterate in Distributed Stochastic Gradient Descent with Momentum
Cheng, Difei
Jin, Ruinan
Qiao, Hong
Zhang, Bo
Optimization and Control
40G15
G.1.0
Distributed stochastic gradient methods are widely used to preserve data privacy and ensure scalability in large-scale learning tasks. While existing theory on distributed momentum Stochastic Gradient Descent (mSGD) mainly focuses on time-averaged convergence, the more practical last-iterate convergence remains underexplored. In this work, we analyze the last-iterate convergence behavior of distributed mSGD in non-convex settings under the classical Robbins-Monro step-size schedule. We prove both almost sure convergence and $L_2$ convergence of the last iterate, and derive convergence rates. We further show that momentum can accelerate early-stage convergence, and provide experiments to support our theory.
title Convergence Analysis of the Last Iterate in Distributed Stochastic Gradient Descent with Momentum
topic Optimization and Control
40G15
G.1.0
url https://arxiv.org/abs/2505.10889