Abstract: Graph neural networks (GNNs) have become the de facto standard for learning over graph-structured data, yet they often falter when labels are scarce, graphs are noisy, or structural patterns ...