Research map / Adversarial robustness and security / Attacks on text and graphs
Interpretable and Robust Behavior Abstraction via Environment-Disentangled Heterogeneous Graph
Zhibin Ni, Hai Wan, Xibin Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add heterogeneous subgraph disentanglement (HSD) (ablation) | EDHGNN · HBGE IM-1 | ROC AUC (%): 62.62 -> 89.52 | Improved |
| add HSD (ablation) | EDHGNN · HBGE Apache-1 | accuracy (%): 91.73 -> 98.45 | Improved |
| add relevance discriminator (RD) (ablation) | EDHGNN · HBGE Apache-1 | ROC AUC (%): 80.10 -> 89.09 | Improved |
| add label-guided graph reconstructor (LGGR) (ablation) | EDHGNN · HBGE IM-1 | accuracy (%): 71.45 -> 75.75 | Improved |
| add adapted graph-level attention (AGLA) instead of readout (ablation) | EDHGNN · HBGE Apache-1 | accuracy (%): 95.51 -> 98.45 | Improved |
| add AGLA instead of readout (ablation) | EDHGNN · HBGE IM-1 | accuracy (%): 74.28 -> 75.75 | Improved |
| raise discriminator weight beta from low value to 3 (ablation) | EDHGNN · HBGE (all four) | accuracy and ROC AUC: Figure 3 only, no numbers in text | Improved |
| raise discriminator weight beta to 5 and 30 (ablation) | EDHGNN · HBGE (all four) | accuracy and ROC AUC: Figure 3 only, no numbers in text | Hurt |
| EDHGNN vs PGIB, interpretability | PGIB · HBGE Apache-2 | ROC AUC (%): 59.12 -> 61.03 | Improved |
| EDHGNN vs GIB under targeted evasion on Apache-1 | GIB · HBGE Apache-1 | average accuracy drop: 19.71 -> 8.76 | Improved |
| EDHGNN vs PGIB under targeted poisoning on IM-1 | PGIB · HBGE IM-1 | average accuracy drop: 13.68 -> 8.99 | Mixed |
Closest papers by meaning
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.85)
- Disentangled Graph Self-supervised Learning for Out-of-Distribution Generalization (ICML 2024 · similarity 0.84)
- Stealthy Yet Effective: Distribution-Preserving Backdoor Attacks on Graph Classification (NeurIPS 2025 · similarity 0.84)
- Clean-Label Graph Backdoor Attack in the Node Classification Task (AAAI 2025 · similarity 0.84)
- Partitioning Message Passing for Graph Fraud Detection (ICLR 2024 · similarity 0.84)
- Disentangling, Amplifying, and Debiasing: Learning Disentangled Representations for Fair Graph Neural Networks (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
