Adaptive Differential Privacy Mechanism in Collaborative Training of Internet of Vehicles

During the last few years, with the development of the Internet-of- Vehicles (IOV), the application of federated learning to the IOV has become possible. Federated learning modifies the manner in which data is transferred, replacing sensitive data with neural network models. However, federated learning still has the potential to leak personal privacy. In this paper, we propose an adaptive differential privacy mechanism appropriate to federated learning in the IOV. The mechanism considers the privacy of the data set participating in each batch of training, the privacy of the local data for the global data, and the subjective requirements of the user for privacy protection. Finally, we calculate an appropriate local privacy budget. The experimental results show that the method proposed in this paper can efficiently improve the performance of the network within the limit of differential privacy.