Homomorphic Encryption-Enabled Federated Transfer Learning for Privacy-Preserving of Internet of Vehicles

Abstract

The swift incorporation of intelligent cars into the Internet of Vehicles (IoV) has contributed to the creation of vast quantities of various and sensitive data, which has caused serious concerns about the privacy and security of data. The conventional centralized learning models tend to face major difficulties associated with loss of privacy, transmission delay, and data handling. In the meantime, the current federated learning (FL) solutions are often unable to be flexible enough when knowledge is transferred into different heterogeneous vehicular settings. In order to overcome these drawbacks, this paper introduces a Privacy-Preserving Federated Transfer Learning (PP-FTL) model that can smoothly combine transfer learning with secure federated model updates based on homomorphic encryption. PP-FTL framework specifically targets data fusion in IoV networks to provide the overall privacy protection and allow the cross-vehicle collaborative learning, even when there is heterogeneous and distributed data. The model uses an aggregation of encrypted weights and a knowledge distillation process that enables successful adaptation between vehicles. Experimental tests based on real-life vehicular network data prove that the suggested PP-FTL framework outperforms the baseline federated learning frameworks in terms of classification, speed of convergence, and preservation of privacy. System attains an average accuracy of 95.7%, a 30% reduction in overheads of communication, and no data leakage, which proves the appropriateness of the system in real-time applications and privacy-conscious IoV applications.