A Resilient and Privacy-Preserving Distributed Learning Framework for Vehicular Intelligence

Abstract

Edge intelligence is becoming a key need for the next generation of cyber-physical systems (CPS), which need to be able to handle low latency, protect data privacy, and be strong against attacks. Connected and selfdriving cars, which are safety-critical CPS applications, need timely, secure, and reliable intelligence to work well in environments that change quickly. But traditional cloud-based learning systems cause communication delays and serious privacy and security problems. This paper presents a secure hierarchical edge-cloud distributed learning framework for vehicle-to-everything (V2X) environments to tackle these challenges. In the proposed system, vehicles use sensor data from their own onboard systems to train local models and only send model updates through V2X communication. First, these updates are collected at nearby multi-access edge computing (MEC) servers. This makes it easier to make decisions quickly and cuts down on communication costs on the backbone before they are synced globally at the cloud layer. A distributed verification system at the edge helps users log in, control access securely, and keep track of model updates in a way that can’t be tampered with to keep trust between all the nodes. Also, local training uses differential privacy to keep sensitive vehicle data safe while still making the model useful. The proposed framework improves the scalability, resilience, and trustworthiness of distributed vehicular intelligence systems by combining hierarchical federated optimization, edge computing, and privacy-aware learning. The results show that structured edge-cloud collaboration can make learning in future edge-enabled transportation platforms safe and reliable.