Research map / Graph learning / GNN architectures
Certifying Graph Neural Networks Against Label and Structure Poisoning
Lukas Gosch, Xichuan Chen, Yan Scholten, Stephan Günnemann
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 pseudo-labels (co-training and self-training) to each label partition (ablation) | partition-based label-flipping certificate · Cora-ML, PubMed (label flipping) | certified accuracy: clean accuracy about 40% -> around 80% (Cora-ML, k=80) | Improved |
| add pseudo-edges via link prediction to each edge partition (ablation) | partition-based structure certificate · Cora-ML, PubMed (structure perturbation, k=800) | certified accuracy | Improved |
| run co-training (label propagation) first, then self-training (ablation) | ST-GPA label generation · Cora-ML, k = 80 (label flipping) | certified accuracy | Improved |
| replace ParWalks with label propagation for co-training (ablation) | ST-GPA · Cora-ML, k = 80 | certified accuracy | No clear effect |
| add a second round of pseudo-edge and pseudo-label generation (ablation) | ST-GPA · PubMed, LF & SP, k = 200 | certified accuracy | Improved |
| use GCN, GAT, or APPNP as base classifier (ablation) | ST-GPA · Cora-ML, PubMed | certified accuracy | No clear effect |
| increase number of partitions k with ST-GPA versus naive partitioning (ablation) | partition aggregation · Cora-ML, PubMed | certified accuracy | Mixed |
| naively partition node labels with many partitions | DPA on graphs · Cora-ML (k = 281) | labeled nodes per partition: average of only 4 labeled nodes with 7 classes | Hurt |
| naively partition edges with more than 50 partitions | DPA on graphs · Cora-ML | subgraph connectivity | Hurt |
Closest papers by meaning
- Robust Graph Condensation via Classification Complexity Mitigation (NeurIPS 2025 · similarity 0.88)
- Clean-Label Graph Backdoor Attack in the Node Classification Task (AAAI 2025 · similarity 0.88)
- Exact Certification of (Graph) Neural Networks Against Label Poisoning (ICLR 2025 · similarity 0.88)
- Partitioning Message Passing for Graph Fraud Detection (ICLR 2024 · similarity 0.87)
- Rethinking Independent Cross-Entropy Loss For Graph-Structured Data (ICML 2024 · similarity 0.87)
- Stealthy Yet Effective: Distribution-Preserving Backdoor Attacks on Graph Classification (NeurIPS 2025 · similarity 0.86)
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.
