Edinburgh Napier University

  We present a secure and efficient federated learning protocol for autonomous vehicles that resists data leaks, redundancy, and adversarial attacks. Our system combines fast zero-knowledge proofs and compressed Bloom filters to verify updates without exposing private data. Compared to traditional approaches, our method reduces proof sizes by 90% (under 10 KB), memory by up to 75%, and maintains accuracy with less than 4% degradation under 30% attack rates. The entire update cycle completes in under 600 ms, making it practical for real-time use in vehicles. This work advances trustworthy AI deployment in dynamic, resource-limited networks.