Multi-Server Verifiable Aggregation for Federated Learning in Securing Industrial IoT
Liutao Zhao, Haoran Xie, Lin Zhong, Yujue Wang · 2024
The paper introduces a novel approach to address secure aggregation in federated learning, utilizing a multi-server model. This demonstrably secure system effectively mitigates the issues of server loss and single point of failure. The protocol established for secure aggregation operates under the common reference string model. It employs an additive homomorphic secret-sharing scheme alongside a homomorphic Chameleon hash function. This combination results in substantial enhancements in performance, particularly in reducing communication and computational expenses. These improvements have been empirically validated through rigorous testing, showcasing the protocol’s efficacy compared to existing alternatives.