Research map / Graph learning / GNN architectures
Be Confident! Towards Trustworthy Graph Neural Networks via Confidence Calibration
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 |
|---|---|---|---|
| add topology-aware CaGCN calibration instead of no calibration | GCN · Cora 20 labels/class | ECE: 0.1347 -> 0.0401 | Helped |
| replace temperature scaling with CaGCN | GAT · Pubmed 20 labels/class | ECE: 0.0656 -> 0.0356 | Helped |
| replace temperature scaling with CaGCN | GCN · CoraFull 20 labels/class | ECE: 0.1013 -> 0.0776 | Helped |
| use CaGCN vs temperature scaling at higher label rate | GCN · Cora 60 labels/class | ECE: 0.0355 -> 0.0376 | Hurt |
| use CaGCN instead of matrix scaling | GCN · Citeseer 60 labels/class | ECE: 0.0521 -> 0.0546 | Hurt |
| calibrate confidence before pseudo-labeling (add CaGCN to self-training) (ablation) | GCN-st · Cora 20 labels/class | accuracy: 82.28 -> 83.11 | Helped |
| add CaGCN calibration to self-training (ablation) | GCN-st · Pubmed 60 labels/class | accuracy: 83.44 -> 84.47 | Helped |
| add CaGCN calibration to self-training with GAT (ablation) | GAT GCN-st · CoraFull 20 labels/class | accuracy: 62.09 -> 65.46 | Helped |
| add CaGCN-st vs standard self-training | GCN · Citeseer 20 labels/class | accuracy: 73.24 -> 74.90 | Helped |
| use CaGCN vs TS in self-training | GCN · Cora 40 labels/class | accuracy: 84.44 -> 84.37 | No clear effect |
| apply CaGCN to reduce total variation of confidence | GCN · Cora | total variation: 240.267 -> 164.651 | Helped |
| matrix scaling on many-class dataset | GCN · CoraFull | ECE | Hurt |
Closest papers by meaning
- Understanding and Mitigating Under-Confidence in GNNs from the Final Layer (NeurIPS 2026 · similarity 0.94)
- Moderate Message Passing Improves Calibration: A Universal Way to Mitigate Confidence Bias in Graph Neural Networks (AAAI 2024 · similarity 0.91)
- Improving GNN Calibration with Discriminative Ability: Insights and Strategies (AAAI 2024 · similarity 0.89)
- SimCalib: Graph Neural Network Calibration Based on Similarity between Nodes (AAAI 2024 · similarity 0.88)
- Accurate and Scalable Estimation of Epistemic Uncertainty for Graph Neural Networks (ICLR 2024 · similarity 0.88)
- Towards Precise Prediction Uncertainty in GNNs: Refining GNNs with Topology-grouping Strategy (AAAI 2025 · 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.