Research map / Graph learning / GNN architectures
Graph Posterior Network: Bayesian Predictive Uncertainty for Node Classification
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 |
|---|---|---|---|
| use GPN with diffused feature pseudo-counts for left-out-class detection | GPN · Left-out classes | Epistemic OOD-AUC-ROC: +12% to +16% versus uncertainty without network effects | Helped |
| use GPN with feature-level uncertainty for Normal feature perturbations | GPN · Normal feature perturbations | OOD-ACC: +16% to +64% compared to baselines | Helped |
| use feature-only epistemic uncertainty for feature OOD detection | GPN · Bernoulli and Normal feature perturbations | OOD-AUC-ROC: at least +15% and +29% | Helped |
| train with diffusion instead of omitting diffusion (ablation) | GPN · CoraML validation split | LOC-EPIS-AUC-ROC: 70.14 ± 0.93 -> 87.34 ± 0.76 | Helped |
| diffuse pseudo-counts during training rather than only at test time (ablation) | GPN · CoraML validation split | ECE: 30.35 ± 0.86 -> 12.77 ± 1.11 | Helped |
| remove density-based epistemic uncertainty and use APPNP (ablation) | GPN · CoraML validation split | LOC-EPIS-AUC-ROC: 87.34 ± 0.76 -> 84.75±0.06 | Hurt |
| diffuse log pseudo-counts instead of pseudo-counts (ablation) | GPN · CoraML validation split | ACC: 81.57 ± 0.18 -> 82.90 ± 0.26 | Helped |
| diffuse log pseudo-counts instead of pseudo-counts (ablation) | GPN · CoraML validation split | LOC-EPIS-AUC-ROC: 87.34 ± 0.76 -> 70.88 ± 1.74 | Hurt |
| diffuse log pseudo-counts instead of pseudo-counts (ablation) | GPN · CoraML validation split | ECE: 12.77 ± 1.11 -> 8.77 ± 0.33 | Helped |
| perturb node features with unit Gaussian noise | GPN · CoraML, feature shifts | ACC: decreases by less than 5% at 80% perturbed nodes; baselines decrease by more than 50% at 20% | Helped |
| randomly perturb graph edges | GPN · Attributed graph edge shifts | aleatoric uncertainty | Helped |
Closest papers by meaning
- Uncertainty Estimation for Heterophilic Graphs Through the Lens of Information Theory (ICML 2025 · similarity 0.90)
- Accurate and Scalable Estimation of Epistemic Uncertainty for Graph Neural Networks (ICLR 2024 · similarity 0.90)
- Uncertainty for Active Learning on Graphs (ICML 2024 · similarity 0.89)
- A Unified Uncertainty Representation for Graph Neural Networks via Doubly-Spectral Stochastic Expansion (NeurIPS 2026 · similarity 0.89)
- Graph Neural Stochastic Diffusion for Estimating Uncertainty in Node Classification (ICML 2024 · similarity 0.89)
- Learning Latent Graph Structures and their Uncertainty (ICML 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.