Research map / Graph learning / GNN architectures
Rethinking pooling in graph neural 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 |
|---|---|---|---|
| cluster the complement graph instead of the original graph for pooling (ablation) | Graclus · NCI1 | accuracy (%): 80.1 -> 80.2 | No clear effect |
| cluster the complement graph (ablation) | Graclus · ZINC | MAE: 0.42 -> 0.43 | No clear effect |
| cluster the complement graph (ablation) | Graclus · PROTEINS | accuracy (%): 72.6 -> 75.3 | Helped |
| replace learned DiffPool assignments with random assignments (ablation) | DiffPool · graph classification benchmarks | accuracy | No clear effect |
| replace GMN assignments with fixed random matrix (ablation) | GMN · NCI1 | accuracy (%): 80.2 -> 80.2 | No clear effect |
| replace MinCutPool assignments with random normal assignments (ablation) | MinCutPool · Superpixels MNIST | accuracy (%): 89.3 -> 82.0 | Hurt |
| replace MinCutPool assignments with random normal assignments (ablation) | MinCutPool · PROTEINS/NCI109/DD | accuracy | No clear effect |
| use a single convolution before first pooling (ablation) | Graclus · NCI1 | accuracy (%): 80.1 -> 76.6 | Hurt |
| use a single convolution before first pooling (ablation) | DiffPool · Superpixels MNIST | accuracy (%): 89.6 -> 66.6 | Hurt |
| use a single convolution before first pooling (ablation) | Graclus · IMDB-B | accuracy (%): 69.9 -> 69.7 | No clear effect |
| increase DiffPool link prediction loss weight from 1 to 1000 (ablation) | DiffPool · ZINC and SMNIST | supervised validation loss | No clear effect |
| increase GMN key update frequency using unsupervised loss (p=2 vs 5) (ablation) | GMN · IMDB-B | validation supervised loss | Hurt |
Closest papers by meaning
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.92)
- Enhancing Graph Classification Robustness with Singular Pooling (NeurIPS 2025 · similarity 0.90)
- SkipPool: Improved Sparse Hierarchical Graph Pooling with Differentiable Exploration (AAAI 2025 · similarity 0.90)
- ENAHPool: The Edge-Node Attention-based Hierarchical Pooling for Graph Neural Networks (ICML 2025 · similarity 0.90)
- MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks (ICLR 2025 · similarity 0.89)
- On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective (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.