Research map / Label-efficient and robust learning / Knowledge distillation
Surrogate as Teacher: Distillation-Guided Graph Poisoning Attack
Xingyu Peng, Ke Xu
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 |
|---|---|---|---|
| use MetaDist instead of strongest baselines (GraD, Metacon) | meta-gradient poisoning attack · Cora/CiteSeer/PolBlogs/Cora-ML, GCN and GAT | victim accuracy (%): Cora GCN: 69.50 -> 67.23 (GraD vs MetaDist) | Improved |
| use Reverse KL loss instead of standard KL (ablation) | MetaDist · CiteSeer (GCN) | victim accuracy (%): 67.35 -> 57.55 | Improved |
| add Online Adaptive Teacher (vs static teacher) (ablation) | MetaDist · Cora (GCN) | victim accuracy (%): 68.37 -> 67.23 | Improved |
| add gradient-based teacher adaptation to online re-evaluation (ablation) | MetaDist · Cora (GCN) | victim accuracy (%): Cora 69.84 -> 67.23; others small gains | Mixed |
| disable teacher gradient adaptation while keeping online re-evaluation (ablation) | MetaDist · CiteSeer (GCN) | victim accuracy (%): 58.98 vs 57.55 (w/o T-Adapt vs full) | No clear effect |
| use MetaDist against defense models | MetaDist · PolBlogs vs ProGNN | victim accuracy (%): 93.43 -> 85.61 (best baseline 92.18) | Improved |
| transfer attack to GAT victim | MetaDist · Cora (GAT) | victim accuracy (%): 79.62 -> 78.59 (Metacon-D vs MetaDist) | Mixed |
Closest papers by meaning
- Rethinking Label Poisoning for GNNs: Pitfalls and Attacks (ICLR 2024 · similarity 0.87)
- Clean-Label Graph Backdoor Attack in the Node Classification Task (AAAI 2025 · similarity 0.85)
- Stealthy Yet Effective: Distribution-Preserving Backdoor Attacks on Graph Classification (NeurIPS 2025 · similarity 0.85)
- Highly Imperceptible Black-Box Graph Injection Attacks with Reinforcement Learning (AAAI 2025 · similarity 0.85)
- Exact Certification of (Graph) Neural Networks Against Label Poisoning (ICLR 2025 · similarity 0.85)
- Grimm: A Plug-and-Play Perturbation Rectifier for Graph Neural Networks Defending Against Poisoning Attacks (AAAI 2025 · similarity 0.85)
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.
