Adaptive graph neural network protection algorithm based on differential privacy

Junjie Yu, Yong Li, Zhandong Liu, Qianren Yang

Abstract: Highlights•Proposed an adaptive differential privacy GNN model for effective privacy protection.•Developed the MSNAP method to enhance GNN’s adaptability to non-uniform data and outliers.•Designed DAS to optimize training efficiency by reducing runtime.•Introduced progressive training to improve both accuracy and privacy protection.