A MAB based Mobility-Aware Client Scheduling Scheme for Heterogeneous Federated Learning within IoV

Federated Learning (FL) has emerged as a promising paradigm for enabling artificial intelligence (AI) applications with stringent latency requirements in the Internet of Vehicles (IoV). However, the inherent mobility of vehicles, leading to limited residence time, dynamic wireless transmissions, and a time-varying candidate set, degrades the FL training efficiency. Moreover, the convergence rate of FL model suffers from the heterogeneity issues among vehicles. Due to constrained spectrum resources and high connecting density, client scheduling is essential for FL to mitigate network burdens and enhance learning performance. In this paper, we propose a client scheduling scheme based on the Multi-Armed Bandit (MAB) framework that selects effective and high-quality clients for the FL within IoV. In detail, we consider a FL-assisted IoV scenario and construct a candidate client pool based on a mobility model, representing moving vehicles on a unidirectional lane. In this context, we formulate an optimization problem to minimize the overall FL training latency while ensuring model accuracy under constrained resources. Considering the low complexity and scalability of MAB, we reformulate the optimization problem and design a reward function that jointly considers the training latency per round and the contribution of local data for reducing the overall latency. To estimate the contribution of local data on global aggregation, we introduce a utility function based on training loss of local datasets in the MAB reward function. Finally, an ε-greedy algorithm is proposed to balance exploration and exploitation within the dynamic and heterogeneous environment. Simulation results demonstrate that our lightweight and effective scheme outperforms existing approaches in reducing training time and accelerating convergence to the target accuracy. The results confirm that the proposed scheme can effectively address dynamicity and heterogeneity challenges under constrained resources of FL within IoV.