Graph diffusion model learning method and device satisfying differential privacy

Abstract

The invention provides a graph diffusion model learning method and device satisfying differential privacy, and the method comprises the steps: training a denoising network in a graph diffusion model based on a plurality of sub-graphs in a held sub-graph set and a plurality of noise adding graphs corresponding to each sub-graph, the de-noising network carries out node feature extraction based on each noise adding graph to obtain each node feature set; obtaining an edge feature set based on the node feature set, and carrying out weighted fusion on a plurality of edge features connected with nodes at two ends of each edge in the corresponding edge feature set based on node attribute similarity information of the noise-added graph by adopting an attention mechanism to generate each fused edge feature so as to obtain an updated edge feature set; and outputting a prediction result of the corresponding sub-graph based on the updated edge feature set, minimizing the prediction result of the corresponding sub-graph and the loss between the sub-graphs or the privacy budget consumption to reach a preset threshold, and obtaining a trained denoising network satisfying differential privacy. According to the invention, graph learning satisfying differential privacy can be realized.

Images