FL-VAD: An Active Defense-Driven Federated Learning Algorithm for Vehicle Privacy Protection

With the rapid advancements in intelligent transportation systems, autonomous driving technologies, and V2X communications, the volume of data generated by vehicles has surged. Effectively leveraging this data for training machine learning models while ensuring privacy protection has become a pressing challenge. Federated learning, as a distributed learning framework, facilitates collaborative model training across multiple parties while preserving data locality. However, privacy risks, particularly those arising from inference attacks, remain a significant concern. In such attacks, adversaries can deduce sensitive information from the model updates shared by participants. This paper proposes a novel proactive defense algorithm, FL-VAD, designed to empower vehicles to autonomously adopt privacy protection strategies that actively mitigate the risk of data leakage. Specifically, vehicles reduce the likelihood of inference attacks and enhance data security through techniques such as data perturbation, model update control, and dynamic defense mechanisms. By integrating V2X technology, vehicles can collaborate more efficiently in the learning process while safeguarding privacy. Experimental results demonstrate that the proposed approach effectively reduces privacy leakage risks, offering robust privacy protection with minimal impact on model performance. This research introduces innovative solutions for enhancing privacy protection in federated learning within the context of intelligent transportation systems and autonomous driving.