Hierarchical Federated Learning Framework for Safe and Secure Connected and Autonomous Vehicles

Fuelled by the availability of more data and computing power, recent breakthroughs in cloud-based artificial intelligence (AI)/machine learning (ML) have transformed various aspects of our lives, including face recognition and medical diagnosis. However, this centralised approach is not suitable for connected and autonomous vehicles (CAVs) due to the reduced capacity of wireless links, long delays in exchanging data with the cloud, limited scalability, and data privacy concerns. To overcome these limits, federated learning (FL) has recently emerged to enable many clients to collaboratively train a model without sharing their local data. While FL reduces the load on wireless links and appeases some of the privacy concerns, it is still facing several open issues related to intermittent wireless connectivity, performance, security, and scalability. These limits are exacerbated for CAVs due to their unique constraints (e.g., high mobility and stringent safety/security requirements). In this context, this proposal will build a hierarchical federated learning framework for CAVs. A hierarchy of tiers (e.g., vehicle, edge, and cloud) is introduced for better scalability, where the tier hosting the learning process is selected depending on the context. The latest features of 5G/6G networks will be leveraged to improve the performance, safety, and cyber resilience of CAVs.