Research map / Graph learning / GNN architectures
How Attentive are Graph 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 |
|---|---|---|---|
| Replace static GAT attention scoring with dynamic GATv2 scoring | GAT · ogbn-arxiv, ogbn-products, ogbn-mag, ogbn-proteins | accuracy / ROC-AUC: 71.59 -> 71.87; 79.04 -> 80.63; 32.20 -> 32.61; 78.63 -> 79.52 | Helped |
| Replace static GAT attention scoring with dynamic GATv2 scoring | GAT · ogbl-collab | Hits@50: 39.32 -> 42.00; 42.37 -> 42.85; 48.10 -> 48.02; 46.63 -> 49.70 | Mixed |
| Replace static GAT attention scoring with dynamic GATv2 scoring | GAT · ogbl-citation2 | MRR: 79.84 -> 80.33; 75.95 -> 80.14 | Helped |
| Replace static GAT attention scoring with dynamic GATv2 scoring | GAT · QM9 | average relative error: +0% -> -11.5% | Helped |
| Replace GAT with GATv2 | GAT · Pubmed | accuracy: 78.1 -> 78.5 | Helped |
| Use 8 attention heads instead of 1 (ablation) | GAT on ogbn-proteins · ogbn-proteins | ROC-AUC: 70.77 -> 78.63 | Helped |
| Use 8 attention heads instead of 1 (ablation) | GATv2 on ogbn-products · ogbn-products | accuracy: 80.63 -> 78.46 | Hurt |
| Use 8 attention heads instead of 1 (ablation) | GATv2 on ogbn-arxiv · ogbn-arxiv | accuracy: 71.78 -> 71.87 | Helped |
| Use 8 attention heads instead of 1 (ablation) | GATv2 on ogbn-mag · ogbn-mag | accuracy: 32.61 -> 32.52 | No clear effect |
| Use 8 attention heads instead of 1 (ablation) | GATv2 on QM9 · QM9 | average relative error: +91.6% -> -11.5% | Helped |
| Use 8 attention heads instead of 1 (ablation) | GAT on QM9 · QM9 | average relative error: +134.1% -> +0% | Helped |
Closest papers by meaning
- GATE: How to Keep Out Intrusive Neighbors (ICML 2024 · similarity 0.88)
- Faithful and Accurate Self-Attention Attribution for Message Passing Neural Networks via the Computation Tree Viewpoint (AAAI 2025 · similarity 0.87)
- HONGAT: Graph Attention Networks in the Presence of High-Order Neighbors (AAAI 2024 · similarity 0.87)
- Graph Attention is Not Always Beneficial: A Theoretical Analysis of Graph Attention Mechanisms via Contextual Stochastic Block Models (ICML 2025 · similarity 0.86)
- A Dynamical Systems-Inspired Pruning Strategy for Addressing Oversmoothing in Graph Attention Networks (ICML 2025 · similarity 0.85)
- KAA: Kolmogorov-Arnold Attention for Enhancing Attentive Graph Neural Networks (ICLR 2025 · similarity 0.85)
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.