Research map / Learning theory and algorithms / Optimal transport
Fast Estimation of Wasserstein Distances via Regression on Sliced Wasserstein Distances
Khai Nguyen, Hai Nguyen, Nhat Ho
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 |
|---|---|---|---|
| use RG regression on sliced Wasserstein distances instead of Wormhole (100 training samples) | Wasserstein approximation · MNIST Point Cloud, ShapeNetV2, MERFISH, scRNA-seq | R^2: MNIST 0.28 -> 0.93; MERFISH -3.6 -> 0.96 | Improved |
| unconstrained instead of constrained RG (ablation) | RG-s · 4 datasets, 100 samples | R^2: MNIST 0.84 -> 0.93; scRNA-seq 1.00 -> 0.99 | Mixed |
| add lower+upper bound predictors (RG-se) rather than only SW and PWD (RG-s) (ablation) | RG · ShapeNetV2 100 samples | R^2: 0.94 -> 0.95 | Improved |
| use optimization-based predictors (Max-SW, Min-SWGG) only: RG-o vs RG-s (ablation) | RG · ShapeNetV2 kNN k=1 | accuracy (%): 82.1 -> 65.1 | Hurt |
| use RG-seo distances instead of single sliced metrics for kNN | ShapeNetV2 classification · ShapeNetV2 kNN k=5 | accuracy (%): 72.5 (EBSW, k=1) -> 83.5 (RG-seo, k=5) | Improved |
| RG instead of Sinkhorn / Linear OT | Wasserstein approximation · ShapeNetV2 | R^2: Linear OT 0.918 -> RG-s 0.942; Sinkhorn -1.549 | Improved |
| RG-s instead of exact Wasserstein for inference | 200 samples (19,900 pairs) · ShapeNetV2 | inference time: 5,030.966s -> 21.408s | Improved |
| replace Wasserstein calls in Wormhole with calibrated RG (RG-Wormhole) | Wormhole · ModelNet40/ShapeNetV2 | training time and quality | Improved |
| train RG on intra-class pairs only, test on inter-class pairs | RG · ShapeNetV2 | R^2: 0.902, 0.846, 0.932, 0.805 vs about 0.95 | Mixed |
| use higher dimension for Gaussian mixtures | RG constrained weight · MoG simulations | weight on lower bound | Mixed |
Closest papers by meaning
- Wasserstein Wormhole: Scalable Optimal Transport Distance with Transformer (ICML 2024 · similarity 0.92)
- Sliced Wasserstein Estimation with Control Variates (ICLR 2024 · similarity 0.90)
- Tree-Sliced Wasserstein Distance with Nonlinear Projection (ICML 2025 · similarity 0.89)
- A New Robust Partial p-Wasserstein-Based Metric for Comparing Distributions (ICML 2024 · similarity 0.89)
- Stereographic Spherical Sliced Wasserstein Distances (ICML 2024 · similarity 0.88)
- Statistical and Computational Guarantees of Kernel Max-Sliced Wasserstein Distances (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.
