Membership inference attack on graph neural networks
Published in 2021 Third IEEE International Conference on Trust, Privacy and Security in Intelligent Systems and Applications (TPS-ISA), 2021
Graph Neural Networks can leak whether specific nodes were part of a model’s training data. This paper studies realistic black-box membership inference settings for GNNs, analyzes why graph structure intensifies leakage, and proposes defenses that reduce the attacker’s success without harming model utility.
Recommended citation: Iyiola E. Olatunji, Wolfgang Nejdl, and Megha Khosla. (2021). "Membership inference attack on graph neural networks." 2021 Third IEEE International Conference on Trust, Privacy and Security in Intelligent Systems and Applications (TPS-ISA).
Download Paper