FedGS: A federated group synchronization framework for heterogeneous data
Zonghang Li, Yihong He · Software Impacts · 2022
FedGS is a federated group synchronization framework implemented on LEAF-MX. It provides a novel client selection strategy and an innovative group collaborative training protocol based on compound step synchronization to make federated learning excellent on non-independent and identically distributed (non-i.i.d.) data, achieving higher accuracy, lower loss, and faster convergence. FedGS is an open-source, readable, and easy-to-use framework.