Ensuring Privacy and Communication Efficiency Through Partial Parameter Exchange in Collaborative Learning for Vehicular Applications

Vehicle-to-Everything (V2X) communication is vital for enhancing road safety, optimizing traffic flow, and reducing environmental impact. With vehicle sensors generating enormous amounts of sensor data for analyzing driver behavior and facilitating safety applications, the adoption of machine learning (ML) solutions has gained significant interest. Collaborative training of ML models tailored for such vehicular environments while preserving privacy has been explored through various approaches, including Federated Learning. This paper proposes the Partial Exchange of Model Parameters, which improves communication resource efficiency and enhances privacy. More specifically, the proposed approach will be applied to well-known collaborative learning techniques such as Federated Learning and Transfer Learning. Although in this paper the focus is on autoencoders, the proposed approach is general. Experimental results demonstrate a significant enhancement in communication efficiency, coupled with maximized privacy during collaborative training.