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. 1906.02351
22
26

On the Convergence of SARAH and Beyond

5 June 2019
Bingcong Li
Meng Ma
G. Giannakis
ArXivPDFHTML
Abstract

The main theme of this work is a unifying algorithm, \textbf{L}oop\textbf{L}ess \textbf{S}ARAH (L2S) for problems formulated as summation of nnn individual loss functions. L2S broadens a recently developed variance reduction method known as SARAH. To find an ϵ\epsilonϵ-accurate solution, L2S enjoys a complexity of O((n+κ)ln⁡(1/ϵ)){\cal O}\big( (n+\kappa) \ln (1/\epsilon)\big)O((n+κ)ln(1/ϵ)) for strongly convex problems. For convex problems, when adopting an nnn-dependent step size, the complexity of L2S is O(n+n/ϵ){\cal O}(n+ \sqrt{n}/\epsilon)O(n+n​/ϵ); while for more frequently adopted nnn-independent step size, the complexity is O(n+n/ϵ){\cal O}(n+ n/\epsilon)O(n+n/ϵ). Distinct from SARAH, our theoretical findings support an nnn-independent step size in convex problems without extra assumptions. For nonconvex problems, the complexity of L2S is O(n+n/ϵ){\cal O}(n+ \sqrt{n}/\epsilon)O(n+n​/ϵ). Our numerical tests on neural networks suggest that L2S can have better generalization properties than SARAH. Along with L2S, our side results include the linear convergence of the last iteration for SARAH in strongly convex problems.

View on arXiv
Comments on this paper