Privacy-Preserving Federated Learning Secure Aggregation Strategy Based on Permissioned Blockchain

In order to solve the problems of privacy-preserving federated learning in which the poisoning behaviour of client nodes as well as the malicious aggregation behaviour of aggregation server nodes lead to the failure of model training, a secure aggregation privacy-preserving federated learning strategy based on permissioned blockchain is proposed. In order to solve the problem of malicious behavior of aggregation server nodes, a trusted aggregation server node selection algorithm is designed to cancel the right of nodes with malicious aggregation behavior to participate in the aggregation server node election. Each node establishes a reputation value, proposes a reward and punishment algorithm based on the reputation value, and establishes a threshold, and nodes with a reputation value lower than the threshold will be refused to participate in the federated learning process to reduce the threat of client node poisoning attack behaviour on model learning.The experimental results show that the scheme is able to ensure secure model aggregation and achieve high model correctness in the presence of malicious aggregation behaviour at 50% of the nodes and poisoning attacks at 40% of the nodes.