Does Black-box Attribute Inference Attacks on Graph Neural Networks Constitute Privacy Risk?
Published in arXiv, 2023
This paper initiates a systematic study of attribute inference attacks against graph neural networks. We vary the adversary’s assumptions and show that, under black-box access, GNNs often do not reveal substantially more information than strong missing-value estimation baselines.
Recommended citation: Iyiola E. Olatunji, Anmar Hizber, Oliver Sihlovec, and Megha Khosla. (2023). "Does Black-box Attribute Inference Attacks on Graph Neural Networks Constitute Privacy Risk?" arXiv.
Download Paper