GJDNet: Robust Graph Neural Networks via Joint Disentangled Learning Against Adversarial Attacks

2026-05-31

AI Summary

Graph neural networks are highly vulnerable to adversarial attacks, primarily because such attacks introduce edges that contradict the graph’s homophily, leading to a mismatch between structure and features and thereby disrupting neighborhood aggregation. To address this issue, this work proposes GJDNet, a novel framework that jointly decouples representation learning and decision-making for the first time. At the representation level, it employs feature-driven soft structural disentanglement and skewness-aware neighbor filtering to mitigate structural-feature mismatch. At the decision level, it introduces a spherical decision boundary (SDB) to enhance intra-class compactness and inter-class separability. The method demonstrates robust performance across diverse graph connectivity patterns and significantly improves adversarial robustness under varying homophily conditions. Experimental results validate its effectiveness in maintaining stable decision boundaries under perturbations.

Abstract

Graph Neural Networks (GNNs) are vulnerable to adversarial attacks, which inherently invert connectivity patterns by introducing disassortative edges in assortative graphs and assortative edges in disassortative graphs. This structural inversion creates structure-feature mismatches that disrupt neighborhood aggregation across different graph types. However, we find that existing defenses are limited, as they either treat neighborhoods as monolithic under fixed assortativity assumptions or rely on standard softmax classifiers that fail to account for perturbation-induced representation shifts. To further exploit this observation, we adopt a robustness perspective that jointly disentangles node representations and decision spaces, isolating perturbation effects while enforcing well-separated decision regions. Based on this principle, we propose Graph Joint Disentanglement Network (GJDNet), a unified framework for robust node classification across diverse graph assortativity regimes. GJDNet enhances robustness at both representation and decision levels: it employs feature-driven soft structural disentanglement with skewness-aware neighbor filtering to suppress perturbation-induced structure-feature mismatches, and introduces a Spherical Decision Boundary (SDB) to promote intra-class compactness and inter-class separation in the embedding space, thereby stabilizing decision boundaries under perturbations. Theoretical analysis provides insights into the effectiveness of the proposed disentangled representation and decision mechanisms, while extensive experiments demonstrate that GJDNet consistently achieves strong robustness across graphs with different connectivity regimes.

Problem

• Graph Neural Networks
• Adversarial Attacks
• Structure-Feature Mismatch
• Assortativity
• Robustness

Innovation

• Graph Neural Networks
• Adversarial Robustness
• Disentangled Learning
• Spherical Decision Boundary
• Assortativity