Federated Learning in IoV

Authored by Xiaoya Hu, Ruiqin Li, Yuqiao Ning, Kaoru Ota, and Licheng Wang

Full title—A Data Sharing Scheme Based on Federated Learning in IoV

As the functions of connected vehicles become more complicated, the amount and type of data they generate during driving are increasing. Sharing the data among connected vehicles can improve users' driving experience and reduce traffic pressure.

However, the leakage of users' private information caused by data sharing may harm the interest of vehicle users and even endanger their lives. Meanwhile, since the quality of data collected by vehicles varies, high-quality data sharing means users get more reliable services, while low-quality data can reduce service reliability, drive experience, or cause traffic accidents.

Therefore, developing an effective and highly reliable privacy protection scheme for shared data becomes a pressing problem to be solved. At the same time, since data sharing requires low latency, it is also a challenge to introduce a privacy protection mechanism without significant delay.

In this paper, we propose a decentralized federated learning-based data sharing scheme that provides strong privacy protection for data and improves the system's robustness. In particular, we decouple the data request process from the data sharing process to improve the sharing efficiency.

Then we propose a TOP-K-based nodes selection scheme to improve the accuracy of the trained models and ensure data reliability. The security analysis shows that the scheme can resist attacks and achieve secure data sharing.

Finally, the scheme's effectiveness and the model prediction's efficiency are verified through experiments. The results show that the scheme has high accuracy, efficiency, and security.

Full Article: IEEE Transactions on Vehicular Technology, Early Access, 2023