Federated Data Enhancement Method for Internet of Vehicles Based on GAN

• 1、Beijing Key Laboratory of Security and Privacy in Intelligent Transportation, Beijing Jiaotong University, Beijing 100044

Abstract：With the wide application of the Internet of Vehicles(IoV), vehicles have entered the "big data era". However, vehicle data contains a large number of user sensitive information, so there is privacy disclosure problem in the process of use. To solve this problem, many scholars have proposed to use federated learning technology. However, the non-independent and identically distributed data of the IoV has brought great challenges to the application of federated learning. Hence, in view of the limited resources of vehicle nodes in the IoV, this paper proposes a data enhancement scheme based on multiple-discriminators Generative Adversarial Networks. In this scheme, the local data is used to train the discriminator and upload the parameters at the vehicle client. The RSU uses the aggregated discriminator to train the generator. The generator generates the data and sends it to the vehicle who selects the data to supplement, so as to generate an independent identically distributed data set. This scheme not only protects the privacy of users\' local data, but also improves the non-independent and identically distributed data and alleviates the problem of limited vehicle customer resources. The experimental results show that this scheme effectively improves the global model accuracy degradation caused by non-independent and identically distributed data, and improves the accuracy rate by 10%~20% without sacrificing data privacy.

Keywords： data security and computer security Internet of Vehicles federated learning Generative Adversarial Networks Non independent and identically distributed privacy protection