Research map / Graph learning / GNN architectures
Learnable Graph Convolutional Attention Networks
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 |
|---|---|---|---|
| convolve features before computing attention score (CAT) instead of plain GAT | GAT on ogbn-arxiv · ogbn-arxiv | test accuracy (%): 71.58 to 72.14 | Helped |
| convolve features before attention score (CAT) instead of plain GAT | GAT on ogbn-products · ogbn-products | test accuracy (%): 78.53 to 77.38 | Hurt |
| learnable interpolation between GCN, GAT and CAT (L-CAT) in place of GAT | GAT across six real datasets · Amazon Computers / Photo / GitHub / PagePage / Physics / TwitchEN | test accuracy (%): +0.75 Computers, +0.91 Photo, +0.74 GitHub, +0.65 PagePage, ~0 Physics, +0.29 Twitch | Helped |
| learnable interpolation (L-CAT) in place of GAT | GAT on OGB datasets · ogbn-proteins / ogbn-mag | AUC-ROC / accuracy: proteins +0.55; mag +0.32 | Mixed |
| GCN as a baseline on real datasets | GCN vs GAT on Amazon Computers and Photo · Amazon Computers, Amazon Photo | test accuracy (%): GCN 90.59 vs GAT 89.59; 95.13 vs 94.02 | Helped |
| learnable interpolation (L-CAT) on GCN-to-GAT layer choice in 4-layer networks, across initializations on proteins | GAT on ogbn-proteins with uniform init, 8 heads · ogbn-proteins 8h uniform | AUC-ROC (%): 78.85 vs 72.23 for GAT | Helped |
| convolve features before score (CAT) in the 1-head proteins setting (ablation) | GAT on ogbn-proteins (1 head, uniform init) · ogbn-proteins 1h uniform | AUC-ROC (%): 64.32 vs 59.73 | Helped |
| convolve features before score (CAT) with 8 heads on proteins normal init (ablation) | GAT on ogbn-proteins (8 heads, normal init) · ogbn-proteins 8h normal | AUC-ROC (%): 74.67 vs 79.08 | Hurt |
Closest papers by meaning
- Graph Attention Networks (ICLR 2018 · similarity 0.92)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.91)
- Towards Deep Attention in Graph Neural Networks: Problems and Remedies (ICML 2023 · similarity 0.91)
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.91)
- Simple and Deep Graph Convolutional Networks (ICML 2020 · similarity 0.90)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (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.