English

Variance Reduced Local SGD with Lower Communication Complexity

Machine Learning 2020-01-01 v1 Distributed, Parallel, and Cluster Computing Optimization and Control Machine Learning

Abstract

To accelerate the training of machine learning models, distributed stochastic gradient descent (SGD) and its variants have been widely adopted, which apply multiple workers in parallel to speed up training. Among them, Local SGD has gained much attention due to its lower communication cost. Nevertheless, when the data distribution on workers is non-identical, Local SGD requires O(T34N34)O(T^{\frac{3}{4}} N^{\frac{3}{4}}) communications to maintain its \emph{linear iteration speedup} property, where TT is the total number of iterations and NN is the number of workers. In this paper, we propose Variance Reduced Local SGD (VRL-SGD) to further reduce the communication complexity. Benefiting from eliminating the dependency on the gradient variance among workers, we theoretically prove that VRL-SGD achieves a \emph{linear iteration speedup} with a lower communication complexity O(T12N32)O(T^{\frac{1}{2}} N^{\frac{3}{2}}) even if workers access non-identical datasets. We conduct experiments on three machine learning tasks, and the experimental results demonstrate that VRL-SGD performs impressively better than Local SGD when the data among workers are quite diverse.

Keywords

Cite

@article{arxiv.1912.12844,
  title  = {Variance Reduced Local SGD with Lower Communication Complexity},
  author = {Xianfeng Liang and Shuheng Shen and Jingchang Liu and Zhen Pan and Enhong Chen and Yifei Cheng},
  journal= {arXiv preprint arXiv:1912.12844},
  year   = {2020}
}

Comments

25 pages, 6 figures. The paper presents a novel variance reduction algorithm for Local SGD

R2 v1 2026-06-23T12:58:47.626Z