Pruning Graphs by Adversarial Robustness Evaluation to Strengthen GNN Defenses
Yongyu Wang
TL;DR
The paper tackles adversarial robustness in graph neural networks by pruning edges through spectral robustness evaluation. It trains GNNs to obtain latent embeddings, models the data manifold with a latent $k$-NN graph, and uses the leading generalized eigenpairs of $L_Y^{+}L_X$ to compute Spade scores that quantify edge non-robustness via $\|V_s^{\top} e_{p,q}\|_2^2$. Edges with high Spade scores are removed to form a pruned graph, which is then used for downstream training. Experiments on CiteSeer show a modest drop in clean accuracy (e.g., from $68.4\%$ to $66.2\%$) but substantial gains in robustness against model-aware structural perturbations, especially in high-perturbation regimes, demonstrating the practical value of spectral edge pruning for robust GNNs. The method provides a principled, scalable approach to identify and discard potentially harmful connections, with implications for deploying GNNs in noisy or adversarial environments.
Abstract
Graph Neural Networks (GNNs) have emerged as a dominant paradigm for learning on graph-structured data, thanks to their ability to jointly exploit node features and relational information encoded in the graph topology. This joint modeling, however, also introduces a critical weakness: perturbations or noise in either the structure or the features can be amplified through message passing, making GNNs highly vulnerable to adversarial attacks and spurious connections. In this work, we introduce a pruning framework that leverages adversarial robustness evaluation to explicitly identify and remove fragile or detrimental components of the graph. By using robustness scores as guidance, our method selectively prunes edges that are most likely to degrade model reliability, thereby yielding cleaner and more resilient graph representations. We instantiate this framework on three representative GNN architectures and conduct extensive experiments on benchmarks. The experimental results show that our approach can significantly enhance the defense capability of GNNs in the high-perturbation regime.