1results about How to "Achieve defense" patented technology

This invention provides a privacy-preserving method, system, and device for perturbation-based graph federation learning. Specifically, it includes the following steps: S0, acquiring original local graph data in the local environment of each federation participant, the original local graph data including graph node feature vectors and a first adjacency matrix; S1, calculating graph node embedding representation vectors based on the graph node feature vectors and the first adjacency matrix, simultaneously calculating the average degree of nodes, and then constructing a node similarity matrix; S2, selecting a potential edge for any node based on the node similarity matrix and average degree generated in S1, generating a potential edge set, and constructing a second adjacency matrix; S3, performing a secondary screening of the potential edges based on the potential edge set and the second adjacency matrix obtained in step S2, constructing a final adjacency matrix; S4, regenerating the graph node embedding representation vectors based on the final adjacency matrix obtained in S3, and uploading them to a central server.