Research map / Graph learning / GNN architectures
Hourglass Persistence for Graphs, Simplices, and Cells
Mattie Ji, Indradyumna Roy, Vikas K Garg
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 contraction schedule to forward-only learned filtration (ablation) | learnable PH (GIN) · NCI1 | accuracy: 76.72 -> 81.27 | Improved |
| add contraction schedule to forward-only learned filtration (ablation) | learnable PH (GIN) · ZINC | MAE: 0.62 -> 0.40 | Improved |
| add contraction schedule to forward-only learned filtration (ablation) | learnable PH (GCN) · ZINC | MAE: 0.86 -> 0.44 | Improved |
| add learned forward filtration to backward-only contraction (ablation) | learnable PH (GIN) · PROTEINS | accuracy: 70.54 -> 73.51 | Improved |
| add contraction to forward filtration (ablation) | learnable PH (GIN) · IMDB-B | accuracy: 74.67 -> 72.00 | Hurt |
| add contraction to forward filtration (ablation) | learnable PH (GCN) · NCI109 | accuracy: 71.91 -> 75.87 | Improved |
| replace PH with full forward-backward learnable persistence | PH (GIN) · PROTEINS | accuracy: 69.35 -> 73.51 | Improved |
| replace PH with full forward-backward learnable persistence | PH (GIN) · OGBG-MOLHIV | AUC: 74.34 -> 72.34 | Hurt |
| replace RePHINE with forward-backward persistence | RePHINE (GCN) · NCI109 | accuracy: 79.50 -> 75.87 | Hurt |
| replace RePHINE with forward-backward persistence | RePHINE (GIN) · NCI1 | accuracy: 78.75 -> 81.27 | Improved |
| add 0D early-connectivity augmentation (RePHINE) vs learned filtration without it (ablation) | Fwd-only/Bwd-only · ZINC and MOLHIV | MAE and AUC: 0.62 -> 0.41 (GIN ZINC MAE) | Improved |
| integrate extended persistence as (f,-f) forward-backward instead of PersLay post-hoc layer | extended persistence · NCI1 etc. | accuracy | Improved |
Closest papers by meaning
- Graph Persistence goes Spectral (NeurIPS 2025 · similarity 0.90)
- Positional Encoding meets Persistent Homology on Graphs (ICML 2025 · similarity 0.90)
- Boosting Graph Pooling with Persistent Homology (NeurIPS 2024 · similarity 0.89)
- On topological descriptors for graph products (NeurIPS 2025 · similarity 0.89)
- Graphcode: Learning from multiparameter persistent homology using graph neural networks (NeurIPS 2024 · similarity 0.89)
- Differentiability and Optimization of Multiparameter Persistent Homology (ICML 2024 · similarity 0.89)
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.
