Research map / Graph learning / GNN architectures
GRAND: Graph Neural Diffusion
Ben Chamberlain, James Rowbottom, Maria I Gorinova, Michael Bronstein, Stefan Webb, Emanuele Rossi
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 |
|---|---|---|---|
| Use GRAND-l instead of GDE | Cora, Planetoid splits · Cora, Planetoid splits | test accuracy: 83.8 \pm 0.5 -> 84.7 \pm 0.6 | Improved |
| Use GRAND-nl-rw instead of GDE | CiteSeer, Planetoid splits · CiteSeer, Planetoid splits | test accuracy: 72.5 \pm 0.5 -> 73.6 \pm 0.3 | Improved |
| Use GRAND-nl-rw instead of CGNN | PubMed, Planetoid splits · PubMed, Planetoid splits | test accuracy: 80.2 \pm 0.3 -> 81.0 \pm 0.4 | Improved |
| Use GRAND-l instead of GAT | Cora, random splits · Cora, random splits | test accuracy: 81.8 \pm 1.3 -> 83.6 \pm 1.0 | Improved |
| Use GRAND-nl-rw instead of GCN | CiteSeer, random splits · CiteSeer, random splits | test accuracy: 71.9 \pm 1.9 -> 74.1 \pm 1.7 | Improved |
| Use GRAND-l instead of GAT | PubMed, random splits · PubMed, random splits | test accuracy: 78.7 \pm 2.3 -> 78.8 \pm 1.7 | No clear effect |
| Use GRAND-l instead of CGNN | Coauthor CS, random splits · Coauthor CS, random splits | test accuracy: 92.3 \pm 0.2 -> 92.9 \pm 0.4 | Improved |
| Use GRAND-nl-rw instead of GAT-ppr | Computer, random splits · Computer, random splits | test accuracy: 85.4 \pm 0.3 -> 85.8 \pm 1.5 | No clear effect |
| Use GRAND-nl-rw instead of GDE | Photo, random splits · Photo, random splits | test accuracy: 92.4 \pm 2.0 -> 92.5 \pm 1.0 | No clear effect |
| Use GRAND-nl-rw instead of GAT | ogb-arxiv, random splits · ogb-arxiv, random splits | test accuracy: 73.65 \pm 0.11 -> 72.23 \pm 0.20 | Hurt |
| Use scaled dot product attention rather than Bahdanau attention (ablation) | GRAND · Node classification benchmarks | test accuracy | Improved |
| Increase adaptive solver tolerances relative to defaults (ablation) | GRAND with adaptive step-size ODE solvers · Node classification benchmarks | prediction accuracy | Improved |
Closest papers by meaning
- GRAND++: Graph Neural Diffusion with A Source Term (ICLR 2022 · similarity 0.94)
- A Generalized Neural Diffusion Framework on Graphs (AAAI 2024 · similarity 0.92)
- Re-Think and Re-Design Graph Neural Networks in Spaces of Continuous Graph Diffusion Functionals (NeurIPS 2023 · similarity 0.91)
- Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.90)
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · similarity 0.89)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 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.
