vix.ing · top · new · best · stats

Tackling the Objective Inconsistency Problem in Heterogeneous Federated Optimization

2020/07/15 by Jianyu Wang, Wang, Jianyu, Qinghua Liu +7 · 575 citations
Computer Science · Mathematics · #Algorithm #Computation #Computer science #Convergence (economics) #Function (biology) #IoT and Edge/Fog Computing #Mathematical optimization #Mathematics #Optimization problem #Point (geometry) #Privacy-Preserving Technologies in Data #Stationary point #Stochastic Gradient Optimization Techniques #Trust region #cs.DC #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.2007.07481

published in arXiv (Cornell University) (Cornell University)

arxiv created 2020/07/15 · openalex publication_date 2020/07/15 · arxiv updated 2020/07/16 · openalex created_date 2020/07/23 · openalex updated_date 2026/08/08

Abstract

In federated optimization, heterogeneity in the clients' local datasets and computation speeds results in large variations in the number of local updates performed by each client in each communication round. Naive weighted aggregation of such models causes objective inconsistency, that is, the global model converges to a stationary point of a mismatched objective function which can be arbitrarily different from the true objective. This paper provides a general framework to analyze the convergence of federated heterogeneous optimization algorithms. It subsumes previously proposed methods such as FedAvg and FedProx and provides the first principled understanding of the solution bias and the convergence slowdown due to objective inconsistency. Using insights from this analysis, we propose FedNova, a normalized averaging method that eliminates objective inconsistency while preserving fast error convergence.

Citations

Cited by

Related