Research map / Graph learning / GNN architectures
ASAP: Adaptive Structure Aware Pooling for Learning Hierarchical Graph Representations
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 ASAP instead of the strongest reported baseline | Graph classification model · D&D | accuracy: 76.45 $\pm$ 0.97 -> 76.87 \pm 0.7 | Helped |
| Use ASAP instead of the strongest reported baseline | Graph classification model · PROTEINS | accuracy: 73.91 $\pm$ 0.72 -> 74.19 \pm 0.79 | Helped |
| Use ASAP instead of the strongest reported baseline | Graph classification model · NCI1 | accuracy: 69.00 $\pm$ 0.49 -> 71.48 \pm 0.42 | Helped |
| Use ASAP instead of the strongest reported baseline | Graph classification model · NCI109 | accuracy: 68.59 $\pm$ 0.67 -> 70.07 \pm 0.55 | Helped |
| Use ASAP instead of the strongest reported baseline | Graph classification model · FRANKENSTEIN | accuracy: 63.44 $\pm$ 0.65 -> 66.26 \pm 0.47 | Helped |
| Add neighborhood node aggregation to both fitness scoring and cluster representation (ablation) | ASAP · FRANKENSTEIN and NCI1 validation | accuracy: 67.4 $\pm$0.6 -> 67.8 $\pm$ 0.6; 69.9 $\pm$ 2.5 -> 70.7 $\pm$ 2.3 | Helped |
| Use M2T attention instead of T2T or S2T (ablation) | ASAP · NCI1 validation | accuracy: 70.3 $\pm$ 2.0 -> 70.7 $\pm$ 2.3; 69.9 $\pm$ 2.0 -> 70.7 $\pm$ 2.3 | Helped |
| Use M2T attention instead of T2T or S2T (ablation) | ASAP · FRANKENSTEIN validation | accuracy: 67.6 $\pm$ 0.5; 67.7 $\pm$ 0.5; 67.8 $\pm$ 0.6 | No clear effect |
| Replace GCN fitness scoring with LEConv (ablation) | ASAP · FRANKENSTEIN validation | accuracy: 62.7$\pm$0.3 -> 67.8$\pm$0.6 | Helped |
| Replace GCN fitness scoring with LEConv (ablation) | ASAP · NCI1 validation | accuracy: 65.4$\pm$2.5 -> 70.7$\pm$2.3 | Helped |
| Use LEConv rather than Basic-LEConv for fitness scoring (ablation) | ASAP · FRANKENSTEIN and NCI1 validation | accuracy: 63.1$\pm$0.7 -> 67.8$\pm$0.6; 69.8$\pm$1.9 -> 70.7$\pm$2.3 | Helped |
| Compute soft edge weights between pooled clusters (ablation) | ASAP · FRANKENSTEIN validation | accuracy: 67.8 $\pm$ 0.6 -> 68.3 $\pm$ 0.5 | Helped |
Closest papers by meaning
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.92)
- SkipPool: Improved Sparse Hierarchical Graph Pooling with Differentiable Exploration (AAAI 2025 · similarity 0.91)
- ENAHPool: The Edge-Node Attention-based Hierarchical Pooling for Graph Neural Networks (ICML 2025 · similarity 0.90)
- Graph Parsing Networks (ICLR 2024 · similarity 0.89)
- MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks (ICLR 2025 · similarity 0.88)
- Enhancing Graph Classification Robustness with Singular Pooling (NeurIPS 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.