Efficient Client Selection in Federated Learning

TL;DR

This paper tackles efficient client selection in federated learning under privacy and system fault constraints. It introduces an adaptive selection framework that jointly optimizes accuracy, privacy, and robustness by selecting top-K clients based on utility, adding Gaussian noise with a privacy budget $\epsilon$, and using checkpointing with interval $t_c^*$ to handle failures. Evaluations on UNSW-NB15 and ROAD for network anomaly detection show a 7% accuracy improvement and 25% reduction in training time over baselines. These results demonstrate a practical, privacy-preserving, and fault-tolerant FL approach with strong performance gains in heterogeneous network environments.

Abstract

Federated Learning (FL) enables decentralized machine learning while preserving data privacy. This paper proposes a novel client selection framework that integrates differential privacy and fault tolerance. The adaptive client selection adjusts the number of clients based on performance and system constraints, with noise added to protect privacy. Evaluated on the UNSW-NB15 and ROAD datasets for network anomaly detection, the method improves accuracy by 7% and reduces training time by 25% compared to baselines. Fault tolerance enhances robustness with minimal performance trade-offs.

Figures (2)

• Figure 1: Performance comparison of the proposed method, ACFL, and FedL2P in terms of accuracy, AUC-ROC, and training time.
• Figure 2: Impact of privacy budgets on accuracy and loss for UNSW-NB15 and ROAD datasets.