ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2006.06377
6
7

STL-SGD: Speeding Up Local SGD with Stagewise Communication Period

11 June 2020
Shuheng Shen
Yifei Cheng
Jingchang Liu
Linli Xu
    LRM
ArXivPDFHTML
Abstract

Distributed parallel stochastic gradient descent algorithms are workhorses for large scale machine learning tasks. Among them, local stochastic gradient descent (Local SGD) has attracted significant attention due to its low communication complexity. Previous studies prove that the communication complexity of Local SGD with a fixed or an adaptive communication period is in the order of O(N32T12)O (N^{\frac{3}{2}} T^{\frac{1}{2}})O(N23​T21​) and O(N34T34)O (N^{\frac{3}{4}} T^{\frac{3}{4}})O(N43​T43​) when the data distributions on clients are identical (IID) or otherwise (Non-IID), where NNN is the number of clients and TTT is the number of iterations. In this paper, to accelerate the convergence by reducing the communication complexity, we propose \textit{ST}agewise \textit{L}ocal \textit{SGD} (STL-SGD), which increases the communication period gradually along with decreasing learning rate. We prove that STL-SGD can keep the same convergence rate and linear speedup as mini-batch SGD. In addition, as the benefit of increasing the communication period, when the objective is strongly convex or satisfies the Polyak-\L ojasiewicz condition, the communication complexity of STL-SGD is O(Nlog⁡T)O (N \log{T})O(NlogT) and O(N12T12)O (N^{\frac{1}{2}} T^{\frac{1}{2}})O(N21​T21​) for the IID case and the Non-IID case respectively, achieving significant improvements over Local SGD. Experiments on both convex and non-convex problems demonstrate the superior performance of STL-SGD.

View on arXiv
Comments on this paper