Design of Federated Learning Engagement Method for Autonomous Vehicle Privacy Protection

An autonomous car is defined as a vehicle that can be operated only by an autonomous driving system and does not require a driver or intervention of a driver or passenger. Various types of data collected for fully autonomous car operation include personal information such as location, image, physical, behavior information, and license number. It is hard for a driver or pedestrian to recognize or control automated collection and processing of personal information by autonomous vehicles in real time. It is, therefore, necessary to prepare and apply appropriate countermeasures such as minimizing the collection of personal information, deleting and encrypting unnecessary information, etc. Federated Learning is a distributed machine learning technique that can learn Al models in cooperation with each other without directly sharing data distributed and stored in multiple locations, such as devices or institutions. Instead of transmitting individual client data to the central server (cloud), the Al model of the central server is sent to the client to train the model with each data. By repeating this process, the global Al model of the central server becomes more general and the accuracy of the local Al model of the client improves. In order to protect personal information collected while driving an autonomous car, this paper proposed a federated learning engagement method, where cross-device was used to engage in the federated learning consisting of a very large number of mobile devices, and based on this, personal information can be protected for the efficient operation of autonomous vehicles.