Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Robustness in Graph Neural Networks: A Hamiltonian Approach
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use a conservative Hamiltonian graph flow instead of conventional graph neural flows | HANG · Cora PGD-GIA, inductive evasion | accuracy (%): 42.81 -> 78.37 | Helped |
| Add Lyapunov-stable quadratic energy design to HANG (ablation) | HANG · Pubmed PGD-GIA, inductive evasion | accuracy (%): 81.81 -> 87.69 | Helped |
| Add Lyapunov-stable quadratic energy design to HANG (ablation) | HANG · Cora PGD-GIA, inductive evasion | accuracy (%): 78.37 -> 79.05; 72.31 -> 71.07 | Mixed |
| Add Lyapunov-stable quadratic energy design to HANG (ablation) | HANG · Polblogs Metattack, perturbation rates 5% and 25% | accuracy (%): 80.19 -> 94.38; 65.80 -> 86.89 | Helped |
| Add Lyapunov-stable quadratic energy design to HANG (ablation) | HANG · Cora Nettack | accuracy (%): 60.12 -> 68.80 | Helped |
| Increase PGD-GIA attack budget | HANG-quad · Cora PGD-GIA, inductive evasion | accuracy (%): 78.41 -> 29.58 | Hurt |
| Increase PGD-GIA attack budget | HANG · Cora PGD-GIA, inductive evasion | accuracy (%): 75.78 -> 66.01 | Hurt |
| Use Hamiltonian conservative flow rather than GRAND | HANG · Cora PGD-GIA, inductive evasion | accuracy (%): 39.65 -> 78.37 | Helped |
| Change the ODE solver among Euler, RK4, Symplectic Euler, and Dopri5 (ablation) | HANG and HANG-quad · Citeseer, clean and PGD-GIA | accuracy (%) | No clear effect |
| Use HANG-quad rather than HANG under white-box PGD-GIA (ablation) | HANG · Citeseer PGD-GIA white-box, inductive evasion | accuracy (%): 67.54 -> 72.21 | Helped |
| Increase MetaGIA attack budget | HANG-quad · Cora MetaGIA, inductive evasion | accuracy (%): 78.10 -> 16.69 | Hurt |
Closest papers by meaning
- Bounding the Expected Robustness of Graph Neural Networks Subject to Node Feature Attacks (ICLR 2024 · similarity 0.90)
- If You Want to Be Robust, Be Wary of Initialization (NeurIPS 2024 · similarity 0.89)
- Robust Graph Neural Networks via Unbiased Aggregation (NeurIPS 2024 · similarity 0.88)
- On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective (NeurIPS 2025 · similarity 0.88)
- Adversarial Robust Generalization of Graph Neural Networks (ICML 2025 · similarity 0.88)
- A Simple and Yet Fairly Effective Defense for Graph Neural Networks (AAAI 2024 · similarity 0.88)
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.