FedCime: An Efficient Federated Learning Approach For Clients in Mobile Edge Computing

Abstract

Federated learning (FL) enables collaborative training of a global model using localized data from multiple devices. However, in resource-constrained mobile edge computing (MEC) environments, non-independent and identically distributed (non-IID) data generated by these devices poses challenges for traditional FL algorithms like Federated Averaging (FedAvg), leading to decreased accuracy of the global model. In addition, dynamic mobile networks with intermittent connectivity, dropouts, and high migration rates hinder the communication of model updates to the central server. To address these challenges, we present FedCime, a novel tier-based FL approach that selects high-utility mobile clients likely to complete training to replace migrating clients during the round of training. Our evaluation shows that FedCime is scalable and significantly improves training performance in terms of accuracy and computational efficiency compared to state-of-the-art FL algorithms.