Research map / Graph learning / GNN architectures
What Makes Graph Neural Networks Miscalibrated?
Hans Hao-Hsun Hsu, Yuesong Shen, Christian Tomani, Daniel Cremers
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 global bias T0 (ablation) | GATS · Cora GCN | ECE (%): 3.72 -> 3.64 | Mixed |
| add global bias T0 (ablation) | GATS · Citeseer GAT | ECE (%): 3.56 -> 3.86 | Hurt |
| add nodewise scaling factors gamma_i (ablation) | GATS · Cora GCN | ECE (%): 3.80 -> 3.64 | Improved |
| add nodewise scaling factors gamma_i (ablation) | GATS · Citeseer GAT | ECE (%): 4.39 -> 3.86 | Improved |
| add relative confidence level (ablation) | GATS · Citeseer GCN | ECE (%): 4.49 -> 4.43 | Improved |
| add relative confidence level (ablation) | GATS · Photo GCN | ECE (%): 1.54 -> 1.51 | No clear effect |
| add attention-based aggregation (ablation) | GATS · Citeseer GAT | ECE (%): 4.81 -> 3.86 | Improved |
| add attention-based aggregation (ablation) | GATS · Cora GCN | ECE (%): 3.63 -> 3.64 | No clear effect |
| sort normalized logits before computing temperature (ablation) | GATS · Cora GCN | ECE (%): 4.35 -> 3.64 | Improved |
| sort logits (ablation) | GATS · Citeseer GCN | ECE (%): 5.87 -> 4.43 | Improved |
| increase attention heads from 1 to 16 (ablation) | GATS with GCN backbone · Cora GCN | ECE (%): 3.95 -> 3.50 | Improved |
| increase attention heads from 8 to 16 (ablation) | GATS with GAT backbone · Citeseer GAT | ECE (%): 3.86 -> 4.20 | Hurt |
Closest papers by meaning
- GETS: Ensemble Temperature Scaling for Calibration in Graph Neural Networks (ICLR 2025 · similarity 0.93)
- HoTS: Homophily-Aware Temperature Scaling for Graph Neural Network Calibration (NeurIPS 2026 · similarity 0.92)
- Towards Precise Prediction Uncertainty in GNNs: Refining GNNs with Topology-grouping Strategy (AAAI 2025 · similarity 0.91)
- SimCalib: Graph Neural Network Calibration Based on Similarity between Nodes (AAAI 2024 · similarity 0.91)
- Understanding and Mitigating Under-Confidence in GNNs from the Final Layer (NeurIPS 2026 · similarity 0.91)
- Improving GNN Calibration with Discriminative Ability: Insights and Strategies (AAAI 2024 · similarity 0.91)
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.
