Secure Federated Learning for Autonomous Vehicles in the NextG Networks

Secure Federated Learning for Autonomous Vehicles in NextG Networks

Principal Investigator: 
Haiying Shen, associate professor, computer science, engineering systems and environment, University of Virginia

Project Description: 
Artificial intelligence combined with fast NextG mobile networks will help incorporate digitization into daily life, including digital twins of real-life objects such as self-driving cars. These connected autonomous vehicles (CAVs) must have fast, easily accessible, and wide-spread AI-based guidance or services to route through traffic jams, maneuver, identify construction, road debris, signage and more. Building accurate machine learning or deep learning (ML/DL) models to provide accurate services or guidance is crucial for autonomous vehicles to be widely adopted. Shifting the computation to the edge of the network, also known as federated learning, offers an ideal solution for ML/DL model training because it uses real-life data from smart phones and vehicles help build the training models. However, federated learning has its drawbacks, including such security vulnerabilities as attackers sending false information or attacking systems. In this project, researchers will study data/model poisoning attacks to help ensure secure, accurate data models to help connected autonomous vehicles safely traverse the roadways.