Privacy-Preserving and Reliable Distributed Federated Learning

Federated learning enables collaborative training of the global model by participants with diverse data sources while preserving data privacy. However, the traditional federated learning architecture faces some challenges, including single-point of server failure and privacy disclosure. To address these challenges, this paper proposes a distributed federated learning scheme based on multi-key homomorphic encryption, which fundamentally solves the problems of server single-point failure and malicious behavior, while effectively protecting the data privacy of participants. The trusted execution environment (TEE) is used to detect the quality of the models and to prevent some malicious participants from executing malicious behavior. Furthermore, an incentive mechanism is designed to encourage participants to actively and honestly perform training tasks. Our scheme satisfies privacy, robustness, and fairness criteria, as demonstrated in our analysis.