Research map / Graph learning / GNN architectures
The expressive power of pooling in Graph Neural Networks
Filippo Maria Bianchi, Veronica Lachi
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 |
|---|---|---|---|
| replace non-expressive Top-k with expressive DMoN pooling | GIN+pooling · EXPWL1 | test accuracy (%): 67.9 -> 99.0 | Improved |
| expressive pooling (DiffPool) instead of no pooling | GIN · EXPWL1 | test accuracy (%): 99.3 -> 97.0 | No clear effect |
| replace Graclus-style regular pooling by Rand-dense random assignment (ablation) | GIN+pooling · EXPWL1 | test accuracy (%): 99.9 -> 91.7 | Hurt |
| replace Graclus by Cmp-Graclus (complement graph) (ablation) | GIN+pooling · EXPWL1 | test accuracy (%): 99.9 -> 91.9 | Hurt |
| random sparse scoring (Rand-sparse) instead of random dense (ablation) | GIN+pooling · EXPWL1 | test accuracy (%): 91.7 -> 62.8 | Hurt |
| global_sum_pool instead of global_max_pool with 3 GIN layers (ablation) | GIN · EXPWL1 | test accuracy (%): 98.3 -> 99.3 | Improved |
| global_sum_pool instead of global_max_pool with 2 GIN layers (ablation) | GIN · EXPWL1 | test accuracy (%): 66.5 -> 92.1 | Improved |
| GIN layers instead of GCN layers, 2 layers sum pool (ablation) | MP layers · EXPWL1 | test accuracy (%): 76.7 -> 92.1 | Improved |
| 3 GIN layers instead of 2 GIN layers before global_max_pool (ablation) | GIN · EXPWL1 | test accuracy (%): 66.5 -> 98.3 | Improved |
| increase pooling ratio from 0.05 to 0.2 for Top-k (ablation) | Top-k · EXPWL1 | test accuracy (%): 69.7 -> 89.4 | Improved |
| increase pooling ratio from 0.05 to 0.2 for SAGPool (ablation) | SAGPool · EXPWL1 | test accuracy (%): 61.6 -> 82.4 | Improved |
| increase pooling ratio from 0.05 to 0.2 for DMoN (expressive) (ablation) | DMoN · EXPWL1 | test accuracy (%): 98.5 -> 98.1 | No clear effect |
Closest papers by meaning
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.90)
- Boosting Graph Pooling with Persistent Homology (NeurIPS 2024 · similarity 0.89)
- ENAHPool: The Edge-Node Attention-based Hierarchical Pooling for Graph Neural Networks (ICML 2025 · similarity 0.89)
- Enhancing Graph Classification Robustness with Singular Pooling (NeurIPS 2025 · similarity 0.89)
- Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness (ICLR 2024 · similarity 0.88)
- Towards Bridging Generalization and Expressivity of Graph Neural Networks (ICLR 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.
