Research map / Graph learning / GNN architectures
Conformal Prediction Sets for Graph Neural Networks
Soroush H. Zargarbashi, Simone Antonelli, Aleksandar Bojchevski
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 1-hop neighborhood score diffusion (DAPS) to APS conformity scores | GCN on CoraML · CoraML transductive | efficiency (set size) and singleton hit ratio: Efficiency -0.50, singleton hit +0.09 vs APS | Improved |
| Add score diffusion to APS (DAPS) on Coauthor CS | GCN · Coauthor CS transductive | singleton hit ratio: +0.13 vs APS | Improved |
| Use RAPS regularization instead of APS | GCN on CoraML · CoraML transductive | singleton hit ratio: -0.22 vs APS | Hurt |
| Add score diffusion to APS on CoraML, plus higher-variance RAPS comparison | GCN · CoraML/PubMed/other small datasets | efficiency: RAPS efficiency ranges from +0.08 to -0.93 vs APS across datasets | Mixed |
| Diffusion with lambda=0 vs tuned lambda (sweep) (ablation) | GCN on CiteSeer · CiteSeer | accuracy: change negligible | No clear effect |
| Use 2-hop diffusion or score propagation (SP) instead of 1-hop diffusion (ablation) | GCN on CoraML · CoraML | efficiency and singleton hit ratio: both variants provide further improvements over 1-hop (qualitative) | Improved |
| Add score diffusion to APS when only hard (one-hot) predictions are available (ablation) | GCN on Coauthor-CS · hard predictions | efficiency and singleton hit ratio: APS fails catastrophically; DAPS remains reasonably close to soft baseline (qualitative) | Improved |
| Add score diffusion on OGBN Arxiv (low homophily) (ablation) | GCN · OGBN Arxiv | efficiency: diffusion does not provide a significant boost (qualitative) | No clear effect |
| Replace NAPS with DAPS when many calibration nodes are unavailable | GCN on CoraML · CoraML transductive, |Vc|=5.2% | applicability and efficiency: NAPS N/A for 79% of nodes; DAPS 1.81 set size applicable to all | Improved |
Closest papers by meaning
- Enhancing Trustworthiness of Graph Neural Networks with Rank-Based Conformal Training (AAAI 2025 · similarity 0.92)
- Valid Conformal Prediction for Dynamic GNNs (ICLR 2025 · similarity 0.91)
- Similarity-Navigated Conformal Prediction for Graph Neural Networks (NeurIPS 2024 · similarity 0.90)
- Almost Surely Asymptotically Constant Graph Neural Networks (NeurIPS 2024 · similarity 0.90)
- GRAPHLCP: Structure-Aware Localized Conformal Prediction on Graphs (NeurIPS 2026 · similarity 0.89)
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.89)
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.
