Research map / Graph learning / GNN architectures
Neural Approximation of Graph Topological Features
Zuoyu Yan, Tengfei Ma, Liangcai Gao, Zhi Tang, Yusu Wang, Chao Chen
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 |
|---|---|---|---|
| predict EPD instead of persistence image directly (GAT vs GAT_PI) (ablation) | GAT · Cora vicinity graphs | PIE: 1.43e-1 -> 2.46e-2 | Improved |
| add min aggregation AGG (GAT vs GAT (+MIN)) (ablation) | GAT · Cora vicinity graphs | W2 distance: 0.655 -> 0.579 | Improved |
| add new edge-feature MSG function (GAT (+MIN) to PDGNN) (ablation) | GAT (+MIN) · Cora vicinity graphs | W2 distance: 0.579 -> 0.241 | Improved |
| remove edge weights from PDGNN (ablation) | PDGNN · Computers vicinity graphs | W2 distance: 0.220 -> 18.63 | Hurt |
| replace true EPD with PDGNN-approximated EPD in PEGN | PEGN · Cora node classification | accuracy: 82.7 -> 82.0 | No clear effect |
| replace true EPD with GIN_PI-predicted persistence image in PEGN | PEGN · Citeseer node classification | accuracy: 71.9 -> 65.7 | Hurt |
| replace true EPD with GIN_PI in TLC-GNN | TLC-GNN · Photo link prediction | AUC-ROC: 98.2 -> 95.8 | Hurt |
| replace true EPD with PDGNN EPD in TLC-GNN | TLC-GNN · Photo link prediction | AUC-ROC: 98.2 -> 98.4 | No clear effect |
| use PDGNN instead of Gudhi for EPD computation on large dense graphs | PDGNN · Computers EPD time | time (s): 8585.50 -> 7.32 | Improved |
| use PDGNN instead of Gudhi on small graphs | PDGNN · Cora EPD time | time (s): 0.44 -> 5.21 | Hurt |
| fine-tune pre-trained PDGNN for one epoch on new graph | PDGNN pretrained on Photo · PubMed | W2 distance: 0.444 -> 0.360 | Improved |
| increase training vicinity graphs from 5% to 80% (ablation) | PDGNN · Cora | W2 distance: 0.391 -> 0.241 | Improved |
Closest papers by meaning
- Graphcode: Learning from multiparameter persistent homology using graph neural networks (NeurIPS 2024 · similarity 0.90)
- Graph Persistence goes Spectral (NeurIPS 2025 · similarity 0.90)
- Boosting Graph Pooling with Persistent Homology (NeurIPS 2024 · similarity 0.88)
- Positional Encoding meets Persistent Homology on Graphs (ICML 2025 · similarity 0.88)
- TopInG: Topologically Interpretable Graph Learning via Persistent Rationale Filtration (ICML 2025 · similarity 0.88)
- TMetaNet: Topological Meta-Learning Framework for Dynamic Link Prediction (ICML 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.
