A Byzantine-Fault-Tolerant Federated Learning Method Using Tree-Decentralized Network and Knowledge Distillation for Internet of Vehicles

Autonomous driving technology achieves self-driving capability through extensive training on user driving data. As data regulations tighten, Federated Learning(FL) helps self-driving in the Internet of Vehicles (IoV) comply and evolve without restrictions. However, this method faces severe structural challenges and susceptibility to parameter pollution attacks. This paper presents TreeChainFL, a Byzantine-Fault-Tolerant FL system that incorporates a blockchain-based decentralized trust mechanism and knowledge distillation(KD) for IoV. This decentralized architecture employs Byzantine fault tolerance to curb the growing threat from malicious nodes and uses KD with meta-learning to fend off poisoning attacks. Our experiments show that TreeChainFL outperforms recent advancements, effectively neutralizing over 50% Byzantine node attacks and ensuring robust structural and computational fault tolerance.