Research map / Learning theory and algorithms / Optimal transport
Online Sinkhorn: Optimal Transport distances from sample streams
Arthur Mensch, 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 |
|---|---|---|---|
| Use Online Sinkhorn instead of batch Sinkhorn | Sinkhorn potential estimation · Gaussian mixtures, 1D and 2D | potential estimation error | Improved |
| Use Online Sinkhorn instead of batch Sinkhorn | Sinkhorn potential estimation · Gaussian mixtures, 10D | potential estimation error | Mixed |
| Use fully-corrective Online Sinkhorn (ablation) | Online Sinkhorn · Gaussian mixtures, 10D | potential estimation error | Improved |
| Use Online Sinkhorn instead of Sinkhorn | Sinkhorn potential estimation · Gaussian mixtures, 1D and 2D, N=10000 | potential estimation error | No clear effect |
| Use randomized Sinkhorn instead of batch Sinkhorn (ablation) | Sinkhorn potential estimation · Gaussian mixtures | potential estimation error | No clear effect |
| Represent potentials with Online Sinkhorn's kernel expansion instead of an RKHS expansion trained by SGD | Potential estimation · Gaussian mixtures, 1D and 2D | potential estimation error in low-mass regions | Improved |
| Use Online Sinkhorn instead of SGD with an RKHS parametrization | Potential estimation · Gaussian mixtures | stability | Improved |
| Warm up Sinkhorn with Online Sinkhorn before standard Sinkhorn updates | Sinkhorn · Discrete GMMs, 3D dragon, and sphere | distance to optimum | Improved |
| Warm up Sinkhorn with Online Sinkhorn before standard Sinkhorn updates | Sinkhorn · Discrete GMMs, 3D dragon, and sphere | convergence speed | Improved |
| Reduce ε while using Online Sinkhorn warmup (ablation) | OS-warmed-up Sinkhorn · Discrete OT datasets | speed gain | Improved |
| Sample without replacement during warmup (ablation) | OS-warmed-up Sinkhorn · Discrete OT datasets | convergence speed | Improved |
| Use fixed batch sizes instead of increasing batch sizes (ablation) | Online Sinkhorn · Continuous OT | potential convergence | Hurt |
Closest papers by meaning
- Screening Sinkhorn Algorithm for Regularized Optimal Transport (NeurIPS 2019 · similarity 0.89)
- Estimation of Stochastic Optimal Transport Maps (NeurIPS 2025 · similarity 0.88)
- Mirror Sinkhorn: Fast Online Optimization on Transport Polytopes (ICML 2023 · similarity 0.88)
- Geometry-Aware Optimal Transport: Fast Intrinsic Dimension and Wasserstein Distance Estimation (NeurIPS 2026 · similarity 0.88)
- Near-linear time approximation algorithms for optimal transport via Sinkhorn iteration (NeurIPS 2017 · similarity 0.88)
- Massively scalable Sinkhorn distances via the Nyström method (NeurIPS 2019 · 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.
