PDTV-FEL: Privacy-preserving and Dual Traceable Verification Federated Edge Learning

Federated edge learning (FEL) emerges as a novel distributed learning paradigm where multiple clients can jointly train a global model without collecting raw data. However, since adversaries can infer sensitive information from the global model and local updates, FEL remains vulnerable to various security challenges in the Internet of Things (IoT). In this paper, we consider two main challenges during the iterative training process: (1) how to ensure the confidentiality of the global model and local updates and (2) how to verify the integrity of the aggregation result and local updates. To address the above challenges, various approaches have been proposed. However, it remains an open problem to ensure the integrity verification of clients and the server while protecting privacy. In this paper, we propose PDTV-FEL, a privacy-preserving and dual traceable verification federated learning scheme. Specifically, we first design a masked MK-CKKS approach that guarantees the confidentiality of the global model and local updates without incurring additional costs. Moreover, we adopt the BLS signature and double trapdoor chameleon hash function for secure traceable verification. The method not only ensures the integrity of local updates and the aggregation result but also enables to identify the wrong phase and epoch in case of incorrect results. Extensive evaluations on various datasets show the efficient verification of PDTV-FEL in comparison to other schemes.