Online Client Scheduling and Resource Allocation for Efficient Federated Edge Learning

Federated learning (FL) enables edge devices to collaboratively train a machine learning model without sharing their raw data. Due to its privacy-preserving advantages, FL has been widely adopted across various real-world applications. However, deploying FL over resource-constrained mobile networks presents challenges, including high training latency and reduced model accuracy, particularly in the presence of data and system heterogeneity. Additionally, mobile networks are affected by communication dynamics, further complicating efficient FL deployment. To address these challenges, this paper investigates optimal client scheduling and resource allocation for FL in resource-constrained, dynamic mobile networks. Specifically, we first analyze how client sampling probability affects model convergence in FL and formulate a stochastic optimization problem to minimize training latency and convergence bound while satisfying the energy constraints. To solve the formulated problem, we propose an online control algorithm, dubbed Lyapunov-based Resource-efficient Online Algorithm (LROA), which operates without requiring the knowledge of future communication dynamics. Extensive experimental results show that the proposed scheme significantly reduces training latency and improves resource efficiency compared to various baselines.