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. 2002.10110
24
40

Revisiting EXTRA for Smooth Distributed Optimization

24 February 2020
Huan Li
Zhouchen Lin
ArXivPDFHTML
Abstract

EXTRA is a popular method for dencentralized distributed optimization and has broad applications. This paper revisits EXTRA. First, we give a sharp complexity analysis for EXTRA with the improved O((Lμ+11−σ2(W))log⁡1ϵ(1−σ2(W)))O\left(\left(\frac{L}{\mu}+\frac{1}{1-\sigma_2(W)}\right)\log\frac{1}{\epsilon(1-\sigma_2(W))}\right)O((μL​+1−σ2​(W)1​)logϵ(1−σ2​(W))1​) communication and computation complexities for μ\muμ-strongly convex and LLL-smooth problems, where σ2(W)\sigma_2(W)σ2​(W) is the second largest singular value of the weight matrix WWW. When the strong convexity is absent, we prove the O((Lϵ+11−σ2(W))log⁡11−σ2(W))O\left(\left(\frac{L}{\epsilon}+\frac{1}{1-\sigma_2(W)}\right)\log\frac{1}{1-\sigma_2(W)}\right)O((ϵL​+1−σ2​(W)1​)log1−σ2​(W)1​) complexities. Then, we use the Catalyst framework to accelerate EXTRA and obtain the O(Lμ(1−σ2(W))log⁡Lμ(1−σ2(W))log⁡1ϵ)O\left(\sqrt{\frac{L}{\mu(1-\sigma_2(W))}}\log\frac{ L}{\mu(1-\sigma_2(W))}\log\frac{1}{\epsilon}\right)O(μ(1−σ2​(W))L​​logμ(1−σ2​(W))L​logϵ1​) communication and computation complexities for strongly convex and smooth problems and the O(Lϵ(1−σ2(W))log⁡1ϵ(1−σ2(W)))O\left(\sqrt{\frac{L}{\epsilon(1-\sigma_2(W))}}\log\frac{1}{\epsilon(1-\sigma_2(W))}\right)O(ϵ(1−σ2​(W))L​​logϵ(1−σ2​(W))1​) complexities for non-strongly convex ones. Our communication complexities of the accelerated EXTRA are only worse by the factors of (log⁡Lμ(1−σ2(W)))\left(\log\frac{L}{\mu(1-\sigma_2(W))}\right)(logμ(1−σ2​(W))L​) and (log⁡1ϵ(1−σ2(W)))\left(\log\frac{1}{\epsilon(1-\sigma_2(W))}\right)(logϵ(1−σ2​(W))1​) from the lower complexity bounds for strongly convex and non-strongly convex problems, respectively.

View on arXiv
Comments on this paper