Research map / Learning theory and algorithms / Optimal transport
Faster Wasserstein Distance Estimation with the Sinkhorn Divergence
Lénaïc Chizat, Pierre Roussillon, Flavien Léger, François-Xavier Vialard, Gabriel Peyré
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 Richardson extrapolation across λ and √2λ | Sinkhorn divergence estimator under a fourth-order approximation assumption · W₂² estimation from random samples | expected estimation error rate: n^{-2/(d'+4)} -> n^{-4/(d'+8)} | Improved |
| Use Sinkhorn divergence instead of the plug-in estimator | W₂² estimation in the d=5 synthetic random-sampling experiment · Smooth compactly supported distributions, d=5 | L₁ error on the first potential | Improved |
| Use Sinkhorn divergence or Richardson extrapolation instead of uncorrected Tλ | Potential estimation on grid-discretized densities · Densities on the 1-D torus | L₁ error on the first potential | Improved |
| Increase λ using the debiased estimator rather than the uncorrected estimator | Potential estimation on synthetic random samples · Smooth compactly supported distributions, d=5 | computational time to reach a target error | Improved |
| Use Richardson extrapolation instead of the Sinkhorn divergence | Potential estimation in the d=5 random-sampling experiment · Smooth compactly supported distributions, d=5 | L₁ error on the first potential | No clear effect |
| Apply the plug-in estimator in place of the Sinkhorn divergence | W₂² estimation for smooth densities · W₂² estimation from random samples | expected estimation error rate: n^{-2/d} versus n^{-2/(d'+4)} | Mixed |
Closest papers by meaning
- Differential Properties of Sinkhorn Approximation for Learning with Wasserstein Distance (NeurIPS 2018 · similarity 0.91)
- Debiased Sinkhorn barycenters (ICML 2020 · similarity 0.91)
- Scalable Approximation Algorithms for $p$-Wasserstein Distance and Its Variants (ICML 2025 · similarity 0.89)
- Accelerating Sinkhorn algorithm with sparse Newton iterations (ICLR 2024 · similarity 0.88)
- Sliced Wasserstein Estimation with Control Variates (ICLR 2024 · similarity 0.88)
- A New Robust Partial p-Wasserstein-Based Metric for Comparing Distributions (ICML 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.
