Research map / Graph learning / GNN architectures
Second-Order Pooling for Graph Neural Networks
Zhengyang Wang, Shuiwang Ji
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 Sum/Avg pooling with SOPool_bimap | GIN-0 · PTC | accuracy (%): 64.6 -> 75.0 | Improved |
| replace Sum/Avg pooling with SOPool_bimap | GIN-0 · MUTAG | accuracy (%): 89.4 -> 95.3 | Improved |
| replace Sum/Avg pooling with SOPool_attn | GIN-0 · PTC | accuracy (%): 64.6 -> 72.9 | Improved |
| replace Sum/Avg pooling with SOPool_bimap | GIN-0 · RDT-B | accuracy (%): 92.4 -> 89.6 | Hurt |
| replace Sum/Avg pooling with SOPool_m_attn | GIN-0 · RDT-M5K | accuracy (%): 57.5 -> 55.8 | Hurt |
| add SOPool_bimap instead of SAGPool as global pooling (ablation) | GIN-0 · NCI1 | accuracy (%): 79.0 -> 83.6 | Improved |
| subtract the mean (CovPool) instead of plain second-order pooling (ablation) | GIN-0 + SOPool_bimap · PTC | accuracy (%): 75.0 -> 73.3 | Hurt |
| subtract the mean (CovPool) instead of plain second-order pooling (ablation) | GIN-0 + SOPool_bimap · IMDB-M | accuracy (%): 54.6 -> 47.8 | Hurt |
| subtract the mean (CovPool) instead of plain second-order pooling (ablation) | GIN-0 + SOPool_bimap · IMDB-B | accuracy (%): 78.4 -> 72.1 | Hurt |
| use first-order softmax attention pooling (AttnPool) instead of second-order attention (ablation) | GIN-0 + SOPool_attn · NCI1 | accuracy (%): 82.8 -> 80.6 | Hurt |
| use AttnPool instead of SOPool_attn (ablation) | GIN-0 · COLLAB | accuracy (%): 81.1 -> 81.8 | Improved |
| use SOPool_m_attn hierarchical pooling instead of DiffPool | hierarchical GNN · PROTEINS | accuracy (%): 68.2 -> 77.1 | Improved |
Closest papers by meaning
- ENAHPool: The Edge-Node Attention-based Hierarchical Pooling for Graph Neural Networks (ICML 2025 · similarity 0.90)
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.90)
- SkipPool: Improved Sparse Hierarchical Graph Pooling with Differentiable Exploration (AAAI 2025 · similarity 0.89)
- Enhancing Graph Classification Robustness with Singular Pooling (NeurIPS 2025 · similarity 0.88)
- Graph Parsing Networks (ICLR 2024 · similarity 0.87)
- Boosting Graph Pooling with Persistent Homology (NeurIPS 2024 · similarity 0.87)
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.
