A Lightweight Privacy-Preserving Asynchronous Federated Learning Scheme in Internet of Vehicles

The Internet of Vehicles (IoV) facilitates wireless communication and information exchange among vehicles, road infrastructure, and pedestrians, creating a comprehensive network for intelligent vehicle control. Within the IoV framework, numerous computing tasks typically transmitted to the cloud for centralized processing lead to significant latency and substantial cloud computing burdens. To optimize IoV efficiency while safeguarding vehicle data privacy, federated learning (FL) has shown promising potential. However, due to vehicle mobility, FL encounters challenges such as communication bandwidth limitations, varying road conditions, and data transmission delays. Additionally, traditional FL methods do not fully ensure the non-disclosure of user data during aggregation. To address these issues, we propose a lightweight encryption-based asynchronous federated learning scheme (LPAsyFL) for privacy protection in IoV. This scheme supports user participation and withdrawal, ensuring security in honest but curious environments. By utilizing lightweight cryptographic primitives and asynchronous aggregation techniques, we introduce a dynamic aggregation mechanism that reduces communication overhead and enhances model aggregation efficiency. Simulation results on various datasets demonstrate that our approach reduces communication costs and improves aggregation efficiency in the vehicular network.