A privacy-preserving federated learning scheme with homomorphic encryption and edge computing

With the rapid advancement of data-driven technologies, safeguarding data privacy has become a focal concern in both academia and industry. Traditional data processing methods typically rely on centralized storage and computation, which increases vulnerability to privacy breaches, particularly during data transmission and storage. To address these challenges, we propose a privacy-preserving federated learning framework integrating homomorphic encryption with an added trust chain. The trust chain enables transparent, immutable recording of data processing stages, significantly enhancing system reliability and trustworthiness. Participants employ homomorphic encryption to ensure that data remains encrypted throughout transmission and aggregation, thereby preventing privacy leaks. Additionally, this method leverages edge computing nodes to boost computational efficiency and reduce communication and computation overhead. Specifically, we designed privacy-preserving algorithm modules for participants, edge nodes, and the sensing platform, including local encryption, secure aggregation, and global update algorithms, ensuring robust data security at every stage. Validation on the MNIST dataset demonstrates that our approach surpasses the traditional federated learning algorithm (FedAvg) in performance. Furthermore, under varying key lengths, the method shows reasonable encryption and decryption overhead, with the trust chain further securing each transaction.