VDFChain: Secure and verifiable decentralized federated learning via committee-based blockchain

Ming Zhou, Zhen Yang, Haiyang Yu, Shui Yu

Abstract: Highlights•We propose the VDFChain, a secure and verifiable decentralized Federated Learning (FL) scheme via committee-based blockchain, which aims to achieve efficient, secure, and correct aggregation models in decentralized FL scenarios.•Unlike traditional methods, we first propose an verifiable approach based on polynomial commitment protocols to defend against Byzantine attacks of committee mechanisms of decentralized FL.•The proposed VDFChain scheme provides secure aggregation by employing mask techniques, thus protecting the privacy of the participant’s gradient, and it also supports participant offline scenarios.•Security analysis shows that our scheme is secure in terms of data privacy preservation and the correctness of aggregation results. Extensive experiments show that our scheme incurs fewer computational and communication overheads when compared with state-of-the-art approaches.