Dynamic Vehicle Selection and Adaptive Aggregation for Asynchronous Federated Learning enabled VANET

The rapid advancement of vehicular networks has paved the way for intelligent transportation systems, offering enhanced traffic management and autonomous driving capabilities. Federated Learning (FL) is emerging as a critical framework that enables the utilization of onboard information and computational resources while protecting data privacy. However, the high mobility of vehicles and the complex nature of wireless channels pose significant challenges for integrating FL into vehicular networks. This work proposes a Dynamic Vehicle Selection and Adaptive Aggregation Asynchronous based Asynchronous Federated Learning (DVSAA-AFL) scheme designed to optimize FL performance in vehicular networks. DVSAA-AFL introduces a novel approach to achieve dynamic vehicle selection corresponding to different conditions and an adaptive aggregation method that adjusts the weights of local models based on various factors. Our study shows the scheme performs better than or consistent with the baseline FL framework.