FedPrune: Personalized and Communication-Efficient Federated Learning on Non-IID Data

Abstract

Federated learning (FL) has been widely deployed in edge computing scenarios. However, FL-related technologies are still facing severe challenges while evolving rapidly. Among them, statistical heterogeneity (i.e., non-IID) seriously hinders the wide deployment of FL. In our work, we propose a new framework for communication-efficient and personalized federated learning, namely FedPrune. More specifically, under the newly proposed FL framework, each client trains a converged model locally to obtain critical parameters and substructure that guide the pruning of the network participating FL. FedPrune is able to achieve high accuracy while greatly reducing communication overhead. Moreover, each client learns a personalized model in FedPrune. Experimental results has demonstrated that FedPrune achieves the best accuracy in image recognition task with varying degrees of reduced communication costs compared to the three baseline methods.