Vehicular networks are rapidly evolving into distributed learning environments, where vehicles continuously generate valuable local data that can enhance collective intelligence, if shared efficiently and securely. This keynote focuses on communication-efficient and privacy-preserving distributed learning frameworks that enable vehicles and Road Side Units to collaboratively train AI models without sharing raw data. By leveraging techniques based on Federated Learning and Gossip-based model exchange, vehicles can adaptively share only the most relevant model updates, reducing bandwidth consumption while preserving privacy. The talk will explore mechanisms for layer-wise update selection and adaptive communication strategies, demonstrating how these techniques balance accuracy and privacy.  Finally, practical use cases such as driver behavior profiling and anomaly detection will be presented to illustrate the potential of the efficient collaborative learning techniques in vehicular networks.
 

暂无