ECSM: An Ensembled Client Selection Mechanism for Efficient Federated Learning

Abstract

This research paper proposes a multi-criteria client selection approach to enhance the efficiency of Federated Learning (FL). While the state-of-the-art client selection in FL mainly focuses on a single characteristic to determine a suitable client for the training process, a multi-criteria selection is needed to provide a more efficient FL system. We introduce the Ensembled Client Selection Mechanism (ECSM) as a novel approach to address this issue. The proposed approach takes into account client accuracy, reputation, and randomness to improve accuracy during the lower communication period. The study employs random client selection to prevent repetitive training and ensure model generalization. The results indicate that the proposed ECSM mechanism can improve FL performance by achieving the desired accuracy with fewer communication rounds. Specifically, the approach improves FL efficiency by 56% when tested on the FMNIST dataset compared to the baseline approach. These findings suggest that the ECSM mechanism can significantly enhance the efficiency of the FL process.