SCAFFOLD: Stochastic Controlled Averaging for Federated Learning

Code

• github.com/TsingZ0/PFL-Non-IID
  pytorch★ 2,091
• github.com/KarhouTam/FL-bench
  pytorch★ 675
• github.com/ramshi236/Accelerated-Federated-Learning-Over-MAC-in-Heterogeneous-Networks
  pytorch★ 61
• github.com/google-research/public-data-in-dpfl
  tf★ 23
• github.com/carbonati/fl-zoo
  pytorch★ 7
• github.com/thejungwon/gc-fed
  pytorch★ 4
• github.com/KarhouTam/SCAFFOLD-PyTorch
  pytorch★ 0
• github.com/ki-ljl/Scaffold-Federated-Learning
  pytorch★ 0

Abstract

Federated Averaging (FedAvg) has emerged as the algorithm of choice for federated learning due to its simplicity and low communication cost. However, in spite of recent research efforts, its performance is not fully understood. We obtain tight convergence rates for FedAvg and prove that it suffers from `client-drift' when the data is heterogeneous (non-iid), resulting in unstable and slow convergence. As a solution, we propose a new algorithm (SCAFFOLD) which uses control variates (variance reduction) to correct for the `client-drift' in its local updates. We prove that SCAFFOLD requires significantly fewer communication rounds and is not affected by data heterogeneity or client sampling. Further, we show that (for quadratics) SCAFFOLD can take advantage of similarity in the client's data yielding even faster convergence. The latter is the first result to quantify the usefulness of local-steps in distributed optimization.

Tasks

Reproductions