A Joint Resource Allocation and Efficient Communication Scheme Based on FL for the Internet of Vehicles

Abstract

Motivated by the rapid development of the Internet of Vehicles (IoV) and fifth-generation mobile communication technology, vehicle clients have put forward higher requirements for reducing communication overhead and enhancing data privacy. The recently emerging distributed learning method, federated learning (FL), can efficiently process data parameters while protecting clients' privacy. Therefore, we propose an FL-empowered resource allocation and communication optimization framework in the IoV. The purpose of this paper is to minimize the loss function of FL while ensuring efficient communication between vehicle devices and the BS. Based on this purpose, the problem of the uplink wireless resource allocation and parameter quantization in the IoV is formulated. Then, the shortest path optimization algorithm is used to achieve wireless resource allocation recursively. Furthermore, an error-feedback enabled quantized stochastic gradient descent algorithm is adopted, which reduces the size of the parameters transmitted to the BS by quantifying the local parameters of the vehicle devices, and the error-feedback mechanism is used to speed up the convergence of the FL. Finally, simulation results show that the proposed scheme outperforms the comparison schemes in terms of global loss and convergence accuracy.