Multi-task Federated Learning based on Client Scheduling in Mobile Edge Computing

Abstract

With the gradual enrichment of mobile edge computing scenarios, more mobile edge smart devices are involved in edge computing. And due to the requirement for data privacy for multi-party machine learning, federated learning has emerged as a secure multi-party machine learning framework. In order to fix the problem of model accuracy due to client data heterogeneity, we propose a multi-task federated learning framework based on client scheduling for mobile edge computing in this paper, which uses client scheduling to reduce the statistical heterogeneity of data and computational redundancy caused by multi-task learning. In addition, an alternating optimization strategy is proposed to solve the unfairness problem of clients in multi-task learning. We conduct a series of experiments and compare them with popular federated learning frameworks, and the results demonstrate that the methods we proposed can significantly improve local model accuracy and solve the client-side fairness problem.