A Federated Learning Framework Based on Blockchain and Adaptive Differential Privacy

Federated learning, as a new paradigm for data sharing, can better ensure the data privacy of all participants. However, it faces two significant challenges: 1) preventing the central server from acting maliciously, and 2) protecting the privacy of model parameters during the federated learning process. To address these issues, this paper proposes A Federated Learning Framework Based on Blockchain and Adaptive Differential Privacy. By replacing the traditional central server with a blockchain, this framework effectively mitigates the risk of a malicious central server. Additionally, it enhances the utility of the model while protecting the privacy of model parameters through the introduction of adaptive differential privacy technology. Theoretical analysis and empirical results demonstrate that our scheme offers superior performance in terms of both security and utility.