The Power of Extrapolation in Federated Learning

Hanmin Li, Kirill Acharya, Peter Richtárik

We propose and study several server-extrapolation strategies for enhancing the theoretical and empirical convergence properties of the popular federated learning optimizer FedProx…