<ul><li>Graph neural networks (GNNs) are vulnerable to membership inference attacks (MIAs) in graph classification tasks.</li><li>Researchers propose a Graph-level Label-Only Membership Inference Attack (GLO-MIA) for GNNs.</li><li>GLO-MIA is based on the stability of the target model's predictions on training versus testing data.</li><li>Evaluation shows GLO-MIA achieves an attack accuracy of up to 0.825, outperforming baseline work.</li></ul>

Graph-Level Label-Only Membership Inference Attack against Graph Neural Networks

Discover more