Toward Byzantine-Resilient Secure AI: A Federated Learning Communication Framework for 6G Consumer Electronics

• Published in: IEEE transactions on consumer electronics 2024-08, Vol.70 (3), p.5719-5728
• Main Authors: Pei, Jiaming, Xue, Rubing, Liu, Chao, Wang, Lukun
• Format: Article
• Language: English
• Subjects: adaptive update strategy; byzantine-robustness; Consumer electronics; Convergence; Data models; Federated learning; Servers; Training; Training data
• ISSN: 0098-3063
• DOI: 10.1109/TCE.2024.3385015

The advent of 6G technology calls for improved communication efficiency and robust security in consumer electronics. Federated learning, while promising for decentralized machine learning, confronts the challenges of ensuring both efficient and secure model training, especially under Byzantine conditions. Traditional methods, which address communication and security separately, fall short when these issues intersect. This research introduces VAUBR, a framework that integrates version-aware parameter updates with Byzantine attack detection to fortify communication in federated learning. VAUBR discerns and prunes anomalous updates, enhancing update efficiency and aggregation security. Our extensive testing shows VAUBR's efficacy in simulations. Yet, its performance in real-world settings with dense user populations and variable communication quality remains to be assessed. Future work will focus on increasing VAUBR's resilience to the latencies of intermittent client connectivity, thereby advancing secure, efficient federated learning for the 6G-powered consumer electronics landscape.