Research map / Graph learning / GNN architectures
High-Order Pooling for Graph Neural Networks with Tensor Decomposition
Chenqing Hua, Guillaume Rabusseau, Jian Tang
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 |
|---|---|---|---|
| add high-order nonlinear CP pooling alongside low-order linear sum interactions (ablation) | tGNN · Cora | accuracy: 0.8623$\pm$0.0107 -> 0.8780$\pm$0.0158 | Improved |
| add high-order nonlinear CP pooling alongside low-order linear sum interactions (ablation) | tGNN · Pubmed | accuracy: 0.8531$\pm$0.0009 -> 0.9018$\pm$0.0015 | Improved |
| add high-order nonlinear CP pooling alongside low-order linear sum interactions (ablation) | tGNN · ZINC | MAE: 0.440$\pm$0.010 -> 0.301$\pm$0.008 | Improved |
| add low-order linear sum interactions to nonlinear CP pooling (ablation) | tGNN · Cora | accuracy: 0.8655$\pm$0.0375 -> 0.8780$\pm$0.0158 | Improved |
| add low-order linear sum interactions to nonlinear CP pooling (ablation) | tGNN · Pubmed | accuracy: 0.8679$\pm$0.0103 -> 0.9018$\pm$0.0015 | Improved |
| add low-order linear sum interactions to nonlinear CP pooling (ablation) | tGNN · ZINC | MAE: 0.407$\pm$0.025 -> 0.301$\pm$0.008 | Improved |
| replace standard GNN pooling with high-order CP pooling and its linear interaction channel | tGNN · Cora | accuracy: 0.8778$\pm$0.0096 -> 0.8808$\pm$0.0131 | Improved |
| replace standard GNN pooling with high-order CP pooling and its linear interaction channel | tGNN · Pubmed | accuracy: 0.8890$\pm$0.0032 -> 0.9080$\pm$0.0018 | Improved |
| replace standard GNN pooling with high-order CP pooling and its linear interaction channel | tGNN · PROTEINS | AUC: 0.8580$\pm$0.0017 -> 0.8255$\pm$0.0049 | Hurt |
| increase CP decomposition rank (ablation) | tGNN · Cora | accuracy: 85.55 -> 87.62 | Improved |
| increase sampled neighbors from 3 to 10 (ablation) | tGNN · Cora | accuracy: 85.55 -> 85.40; 86.25 -> 86.35; 86.06 -> 86.83; 86.76 -> 86.77; 86.97 -> 87.42; 87.33 -> 87.22; 87.62 -> 87.87 | Mixed |
Closest papers by meaning
- Accurate Learning of Graph Representations with Graph Multiset Pooling (ICLR 2021 · similarity 0.90)
- Rethinking pooling in graph neural networks (NeurIPS 2020 · similarity 0.89)
- From Relational Pooling to Subgraph GNNs: A Universal Framework for More Expressive Graph Neural Networks (ICML 2023 · similarity 0.89)
- ASAP: Adaptive Structure Aware Pooling for Learning Hierarchical Graph Representations (AAAI 2020 · similarity 0.89)
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.89)
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 2024 · 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.
