I. Introduction
In The current era of the Internet of Vehicles (IoVs), smart vehicles have the ability to sense a massive amount of environmental data. With this advanced capacity for data processing, smart vehicles can perform Machine Learning (ML) tasks and train ML models with fusion centers, e.g., cloud and edge servers installed at the Base Stations (BSs) or Roadside Units (RSUs). Collaborative ML enables various transportation-related applications, e.g., traffic prediction, road condition analysis, and route planning [1], [2]. In traditional ML algorithms, vehicles need to share the raw data with fusion centers to perform model training, which raises concerns about privacy leakage as the raw data may be eavesdropped during transmission. Especially in IoVs, the vehicular sensitive data such as coordinates, speed and driving preference is closely related to personal safety and traffic condition.