Research map / Graph learning / GNN architectures
Self-Attention Graph Pooling
Junhyun Lee, Inyeop Lee, Jaewoo Kang
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 self-attention graph pooling instead of gPool | Hierarchical graph classification · D&D | accuracy: 75.01 ± 0.86 -> 76.45 ± 0.97 | Improved |
| Use self-attention graph pooling instead of gPool | Hierarchical graph classification · PROTEINS | accuracy: 71.10 ± 0.90 -> 71.86 ± 0.97 | Improved |
| Use self-attention graph pooling instead of DiffPool | Hierarchical graph classification · NCI1 | accuracy: 62.32 ± 1.90 -> 67.45 ± 1.11 | Improved |
| Use self-attention graph pooling instead of DiffPool | Hierarchical graph classification · NCI109 | accuracy: 61.98 ± 1.98 -> 67.86 ± 1.41 | Improved |
| Use self-attention graph pooling instead of gPool | Hierarchical graph classification · FRANKENSTEIN | accuracy: 61.46 ± 0.84 -> 61.73 ± 0.76 | No clear effect |
| Use self-attention graph pooling instead of SortPool | Global graph classification · D&D | accuracy: 72.53 ± 1.19 -> 76.19 ± 0.94 | Improved |
| Use self-attention graph pooling instead of SortPool | Global graph classification · PROTEINS | accuracy: 66.72 ± 3.56 -> 70.04 ± 1.47 | Improved |
| Use self-attention graph pooling instead of SortPool | Global graph classification · NCI1 | accuracy: 73.82 ± 0.96 -> 74.18 ± 1.20 | No clear effect |
| Use self-attention graph pooling instead of SortPool | Global graph classification · NCI109 | accuracy: 74.02 ± 1.18 -> 74.06 ± 0.78 | No clear effect |
| Use self-attention graph pooling instead of SortPool | Global graph classification · FRANKENSTEIN | accuracy: 60.61 ± 0.77 -> 62.57 ± 0.60 | Improved |
| Add two-hop edges using adjacency augmentation (ablation) | SAGPool_h · D&D | accuracy: 76.45 ± 0.97 -> 77.07 ± 0.82 | Improved |
| Add two-hop edges using adjacency augmentation (ablation) | SAGPool_h · PROTEINS | accuracy: 71.86 ± 0.97 -> 71.82 ± 0.81 | No clear effect |
Closest papers by meaning
- Graph Multi-Convolution and Attention Pooling for Graph Classification (TPAMI 2024 · similarity 0.91)
- Rethinking pooling in graph neural networks (NeurIPS 2020 · similarity 0.90)
- AttPool: Towards Hierarchical Feature Representation in Graph Convolutional Networks via Attention Mechanism (ICCV 2019 · similarity 0.89)
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.88)
- ENAHPool: The Edge-Node Attention-based Hierarchical Pooling for Graph Neural Networks (ICML 2025 · similarity 0.87)
- Enhancing Graph Classification Robustness with Singular Pooling (NeurIPS 2025 · similarity 0.86)
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.
