Clustered Federated Learning for Heterogeneous Data (Student Abstract)

Xue Yu, Ziyi Liu, Yifan Sun, Wu Wang

Abstract: Federated Learning (FL) aims to achieve a global model via aggregating models from all devices. However, it can diverge when the data on the users’ devices are heterogeneous. To address this issue, we propose a novel clustered FL method (FPFC) based on a nonconvex pairwise fusion penalty. FPFC can automatically identify clusters without prior knowledge of the number of clusters and the set of devices in each cluster. Our method is implemented in parallel, updates only a subset of devices at each communication round, and allows each participating device to perform inexact computation. We also provide convergence guarantees of FPFC for general nonconvex losses. Experiment results demonstrate the advantages of FPFC over existing methods.