Research map / Learning theory and algorithms / Prediction and decision losses
Persistent Homology With Improved Locality Information for More Effective Delineation
Doruk Öner, Adélie Garin, Mateusz Koziński, Kathryn P. Hess, Pascal Fua
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 thresholding filtration with combined threshold plus height-function filtration | PH-based topological loss (Homo-Reg) · Massachusetts (validation) | APLS: 65.0 -> 68.7 | Improved |
| same change, Betti error | Homo-Reg · Massachusetts (validation) | Betti error: 1.09 -> 0.90 | Improved |
| same change on RTracer | Homo-Reg · RTracer (test) | APLS: 69.9 -> 73.8 | Improved |
| same change on RTracer, Betti error | Homo-Reg · RTracer (test) | Betti error: 1.07 -> 0.89 | Improved |
| same change on 3D Neurons | Homo-Reg · Neurons (3D, test) | APLS: 85.4 -> 86.9 | Improved |
| same change on 3D Brain | Homo-Reg · Brain (3D, test) | APLS: 70.5 -> 73.4 | Improved |
| add PH-based topological loss (with our filtration) to U-Net MSE training | U-Net · RTracer (test) | Betti error: 2.99 -> 0.89 | Improved |
| add PH topological loss on distance maps instead of classification maps | Homo-Pre · Massachusetts (validation) | APLS: 62.5 -> 65.0 | Improved |
| use ConnLoss instead of our PH loss | delineation network · RTracer (test) | APLS: 73.8 -> 75.4 | Hurt |
| use ConnLoss instead of our PH loss, Betti error | delineation network · RTracer (test) | Betti error: 0.89 -> 1.30 | Hurt |
| use new filtration when injecting synthetic errors (ablation) | topological loss term C · synthetic distance maps | probability loss decreases after injecting an error: 0.4 -> 0.2 | Improved |
Closest papers by meaning
- Adaptive Topological Feature via Persistent Homology: Filtration Learning for Point Clouds (NeurIPS 2023 · similarity 0.91)
- Hourglass Persistence for Graphs, Simplices, and Cells (ICLR 2026 · similarity 0.90)
- The Flood Complex: Large-Scale Persistent Homology on Millions of Points (NeurIPS 2025 · similarity 0.88)
- Graphcode: Learning from multiparameter persistent homology using graph neural networks (NeurIPS 2024 · similarity 0.87)
- Differentiability and Optimization of Multiparameter Persistent Homology (ICML 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.
