Game theory-based vehicle selection and channel scheduling for federated learning in vehicular edge networks

Abstract

In the last decade, Intelligent Transportation System (ITS) has benefited from the rapid development of advanced computing and communication technology. From the perspective of the service operator, connected-vehicles are encouraged to participate in model training to improve the intelligence of vehicular edge networks. However, the data exchanges without proper management will raise the users’ privacy concern. In addition, multi-users participating in the model training require an efficient and distributed mechanism to avoid the waste of operation resources. Therefore, it is valuable to study how to select appropriate vehicles to facilitate the model training and data transmission while protecting users’ privacy. In this paper, the vehicular edge network is built with RSU as the task publishers and users as participants. We contextualize Federated Learning in vehicular edge networks with multi-channel transmission. Therefore, it is valuable to study how to select appropriate vehicles to facilitate the model training and data transmission while maintaining the operation efficiency. In this paper, we contextualize Federated Learning in vehicular edge networks with multi-channel transmission. A vehicle selection strategy based on Stackelberg game is designed to identify the vehicles participating in the model training. Furthermore, a sub-channel scheduling strategy is proposed based on Chaos Game Optimization (CGO) for efficient data transmission. Finally, the simulation verifies the service efficiency and operation effectiveness of the proposed strategies in terms of the operating costs, model accuracy and loss.