English

Accelerating Gossip SGD with Periodic Global Averaging

Machine Learning 2021-05-20 v1 Distributed, Parallel, and Cluster Computing

Abstract

Communication overhead hinders the scalability of large-scale distributed training. Gossip SGD, where each node averages only with its neighbors, is more communication-efficient than the prevalent parallel SGD. However, its convergence rate is reversely proportional to quantity 1β1-\beta which measures the network connectivity. On large and sparse networks where 1β01-\beta \to 0, Gossip SGD requires more iterations to converge, which offsets against its communication benefit. This paper introduces Gossip-PGA, which adds Periodic Global Averaging into Gossip SGD. Its transient stage, i.e., the iterations required to reach asymptotic linear speedup stage, improves from Ω(β4n3/(1β)4)\Omega(\beta^4 n^3/(1-\beta)^4) to Ω(β4n3H4)\Omega(\beta^4 n^3 H^4) for non-convex problems. The influence of network topology in Gossip-PGA can be controlled by the averaging period HH. Its transient-stage complexity is also superior to Local SGD which has order Ω(n3H4)\Omega(n^3 H^4). Empirical results of large-scale training on image classification (ResNet50) and language modeling (BERT) validate our theoretical findings.

Keywords

Cite

@article{arxiv.2105.09080,
  title  = {Accelerating Gossip SGD with Periodic Global Averaging},
  author = {Yiming Chen and Kun Yuan and Yingya Zhang and Pan Pan and Yinghui Xu and Wotao Yin},
  journal= {arXiv preprint arXiv:2105.09080},
  year   = {2021}
}

Comments

Accepted to ICML 2021

R2 v1 2026-06-24T02:15:35.670Z