Research map / Learning theory and algorithms / Optimal transport
An Efficient SE(p)-Invariant Transport Metric Driven by Polar Transport Discrepancy-based Representation
Junyi Lin, Dunyao Xue, Jun Yu, Hongteng Xu, Cheng Meng
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 SEINT regularization (alpha=0.3) to training loss | UniGEM · QM9 pre-training | molecular stability (%): 89.4 -> 93.5 | Improved |
| add SEINT regularization (alpha=0.3) to training loss | EDM · QM9 pre-training | molecular stability (%): 87.2 -> 91.5 | Improved |
| add SEINT regularization (alpha=0.3) to training loss | EDM · QM9 pre-training | validity (%): 93.4 -> 96.5 | Improved |
| add SEINT regularization (alpha=0.3) to training loss | UniGEM · QM9 pre-training | validity x uniqueness (%): 92.8 -> 91.8 | Hurt |
| replace L2-Norm coordinate loss with SEINT regularizer (alpha=0.1) (ablation) | UniGEM · QM9 pre-training | molecular stability (%): 89.0 -> 91.0 | Improved |
| increase SEINT proportion from 0.1 to 0.3 (ablation) | UniGEM · QM9 pre-training | molecular stability (%): 91.0 -> 93.5 | Improved |
| add L2-Norm coordinate regularizer (weight 0.2) (ablation) | UniGEM · QM9 pre-training | molecular stability (%): 89.4 -> 85.3 | No clear effect |
| add SEINT regularizer (alpha=0.3) in fine-tuning | UniGEM · QM9 fine-tuning | molecular stability (%): 89.4 -> 94.9 | Improved |
| add SEINT regularizer (alpha=0.3) in fine-tuning | UniGEM · GEOM-Drugs fine-tuning | molecular stability (%): 1.2 -> 13.0 | Improved |
| add SEINT regularizer (alpha=0.3) in fine-tuning | UniGEM · GEOM-Drugs fine-tuning | validity (%): 98.3 -> 97.7 | Hurt |
| use SEINT instead of RISGW for rotation-invariant classification | kNN on distances · ModelNet40-SE(3) | accuracy k=1: 74.9 -> 100.0 | Improved |
| use SEINT instead of GW for runtime | distance computation · ModelNet40-SE(3) | CPU time (h): 2655.9 -> 9.01 | Improved |
Closest papers by meaning
- Wasserstein Wormhole: Scalable Optimal Transport Distance with Transformer (ICML 2024 · similarity 0.86)
- SymDrift: One-Shot Generative Modeling under Symmetries (NeurIPS 2026 · similarity 0.84)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.84)
- Geometry-Aware Optimal Transport: Fast Intrinsic Dimension and Wasserstein Distance Estimation (NeurIPS 2026 · similarity 0.84)
- SE(3)-Stochastic Flow Matching for Protein Backbone Generation (ICLR 2024 · similarity 0.84)
- Joint Metric Space Embedding by Unbalanced Optimal Transport with Gromov–Wasserstein Marginal Penalization (ICML 2025 · similarity 0.84)
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.
