Grouped Federated Learning: A Decentralized Learning Framework with Low Latency for Heterogeneous Devices

In recent years, federated learning (FL) plays an important role in data privacy-sensitive scenarios to perform learning works collectively without data exchange. However, due to the centralized model aggregation for heterogeneous devices in FL, the convergence is delayed by the last updated model after local training, which increases the economic cost and dampens clients' motivations for participating FL. In this paper, we propose a decentralized FL framework by grouping the clients with the similar computing and communication performance, named federated averaging-inspired group-based federated learning (FGFL). Specifically, we provide a cost function and a greedy-based grouping strategy, which divides the clients into several groups to accelerate the convergence of the FL model. The simulation results verify the effectiveness of FGFL for accelerating the convergence of FL with heterogeneous clients. Besides the exemplified convolutional neural network (CNN), FGFL is also applicable with other learning models.