Latency-Constrained Robust Federated Learning for 6G V2X Edge-to-Cloud Networks

Abstract

The integration of Federated Learning (FL) within 6G-enabled Vehicular Ad Hoc Networks (VANETs) holds significant potential for decentralized, privacy-preserving intrusion detection. However, providing Ultra-Reliable Low-Latency Communication (URLLC) while simultaneously providing defense against adversarial poisoning is a major challenge. In this paper, we propose a new framework, namely UA-RFA, based on Uncertainty-Aware Robust Federated Aggregator, which is specifically designed for the 6G edge-to-cloud continuum. As shown by our experimental evaluation, our proposed framework meets stringent Ultra-Reliable Low-Latency Communication requirements, as all vehicular updates are completed within a 20 ms window, with a mean of 13.4 ms. Although our architecture shows significant stability and robustness against adversarial updates, our experiment shows a significant phenomenon of a ‘majority class plateau’ for highly imbalanced network intrusion detection system (NIDS) datasets, resulting in a stable global accuracy of 87.3%.