Efficient Private Set Intersection for Vertical Federated Learning in IoV

With the development of various observation and monitoring devices, a significant volume of heterogeneous data has been collected from electronic components in the Internet of Vehicles (IoV). Traditional machine learning collects heterogeneous datasets for model training, which poses challenges in terms of data security and communication efficiency. As a branch of Federated Learning, Vertical Federated Learning (VFL) is well-suited for the environment of heterogeneous data in IoV, where it transmits model updates instead of the raw dataset, thereby enhancing data security and communication efficiency. Before conducting model training, it is inevitably required to perform Private Set Intersection (PSI) in VFL. Therefore, efficient PSI schemes are advantageous in enhancing the efficiency of tasks. In this article, the proposed scheme combines the oblivious transfer protocol and cuckoo hashing to achieve rapid intersection computation. Compared to the others, this scheme features shorter execution times, smaller space complexity, and comparable model performance.