DPRO-GNN: Bridging differential privacy and advanced optimization for privacy-preserving graph learning

Open Webpage

Yanan Bai, Liji Xiao, Hongbo Zhao, Xiaoyu Shi

Published: 01 Jan 2026, Last Modified: 10 Nov 2025Inf. Sci. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Proposed DPRO-GNN, a privacy-preserving GNN with advanced optimization.•Introduced DP-RangerBC optimizer to correct DP noise bias during training.•Achieved faster convergence and higher accuracy under strict DP limits.•Provided theoretical analysis and experiments proving privacy and performance gains.