Research map / Learning theory and algorithms / Optimal transport
The Wasserstein Transform
Facundo Memoli, Zane Smith, Zhengchao Wan
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 |
|---|---|---|---|
| Iterate WT 3 times instead of 1 at eps=0.8 (ablation) | WT · Grassmann 3nn | error rate: 1.4912 -> 1.0039 | Improved |
| Iterate WT 3 times at eps=0.9 (ablation) | WT · Grassmann 3nn | error rate: 1.2031 -> 0.9753 | Improved |
| Iterate WT further at eps=1.0 (ablation) | WT · Grassmann 3nn | error rate: 1.2328 -> 8.7605 | Hurt |
| Increase eps from 0.9 to 1.2 for WT (ablation) | WT · Grassmann 3nn | error rate: 1.2031 -> 2.5633 | Hurt |
| Replace mean shift (Gaussian kernel) with WT at eps=0.9 (ablation) | WT vs MS · Grassmann 3nn | error rate: 1.9155 -> 1.2031 | Improved |
| Iterate mean shift (Gaussian kernel) further at eps=0.8 (ablation) | MS · Grassmann 3nn | error rate: 1.8148 -> 8.3543 | Hurt |
| Mean shift with truncation kernel instead of Gaussian kernel (ablation) | MS · Grassmann 3nn | error rate: 1.8148 -> 13.1189 | Hurt |
| Apply WT to manifold distance instead of raw distance at eps=0.8 | WT · Grassmann 3nn | error rate: 1.7903 -> 1.0039 | Improved |
| Use WT instead of tangent distance at knn=1 | WT-Exact-1 · MNIST 5K test | misclassified images: 133 -> 121 | Improved |
| Use WT instead of tangent distance at knn=3 | WT-Sinkhorn-1 · MNIST 5K test | misclassified images: 130 -> 118 | Improved |
| Sinkhorn approximation instead of exact OT (ablation) | WT · MNIST 5K test | misclassified images: 121 -> 125 | No clear effect |
| WT instead of mean shift (truncation kernel) | WT · MNIST 5K test | misclassified images: 116 -> 117 | No clear effect |
Closest papers by meaning
- Slicing Wasserstein over Wasserstein via Functional Optimal Transport (ICLR 2026 · similarity 0.88)
- Flowing Datasets with Wasserstein over Wasserstein Gradient Flows (ICML 2025 · similarity 0.86)
- Tree-Sliced Wasserstein Distance with Nonlinear Projection (ICML 2025 · similarity 0.86)
- Statistical and Computational Guarantees of Kernel Max-Sliced Wasserstein Distances (ICML 2025 · similarity 0.86)
- Towards Understanding Gradient Dynamics of the Sliced-Wasserstein Distance via Critical Point Analysis (ICML 2025 · similarity 0.86)
- Fourier Sliced-Wasserstein Embedding for Multisets and Measures (ICLR 2025 · similarity 0.86)
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.
