Research map / Learning theory and algorithms / Optimal transport
Sinkhorn and entropic OT algorithms
Analyze fast solvers for entropic and unbalanced optimal transport.
69 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Optimal transport, keyword cluster “optimal transport · transport ot · sinkhorn”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
15 papers in 2023–24 and 30 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 17.9% to 21.1%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · Near-linear time approximation algorithms for optimal transport via Sinkhorn iteration (NeurIPS 2017 · 292 citations)
- 2019 · On Efficient Optimal Transport: An Analysis of Greedy and Accelerated Mirror Descent Algorithms (ICML 2019 · 57 citations)
- 2020 · On Unbalanced Optimal Transport: An Analysis of Sinkhorn Algorithm (ICML 2020 · 24 citations)
- 2021 · Hilbert Sinkhorn Divergence for Optimal Transport (CVPR 2021 · 9 citations)
- 2022 · Exact Optimal Accelerated Complexity for Fixed-Point Iterations (ICML 2022 · 3 citations)
- 2023 · Re-Basin via Implicit Sinkhorn Differentiation (CVPR 2023 · 10 citations)
- 2024 · Optimal Transport with Cyclic Symmetry (AAAI 2024 · 2 citations)
- 2025 · Optimal Transport Barycenter via Nonconvex-Concave Minimax Optimization (ICML 2025 · 0 citations)
- 2026 · UOTIP: Unbalanced Optimal Transport Map for Unpaired Inverse Problems (ICML 2026 · 0 citations)
Earliest papers here
- Near-linear time approximation algorithms for optimal transport via Sinkhorn iteration (NeurIPS 2017 · 292 citations)
- Tsallis Regularized Optimal Transport and Ecological Inference (AAAI 2017 · 30 citations)
- On Efficient Optimal Transport: An Analysis of Greedy and Accelerated Mirror Descent Algorithms (ICML 2019 · 57 citations)
What has been tried
37 papers here report 264 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 9%, hurt 19%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Replace sparse Hessian approximation with the full Hessian (remove sparsification) | Random assignment, n=2000, Newton stage · Time per iteration (s) | 0.30 -> 11.86 | Hurt | Accelerating Sinkhorn algorithm with sparse Newton iterations (ICLR 2024 · 1 citation) |
| vary the value-iteration update count m | 4-rooms · convergence rate to the true distance | No clear effect | Bisimulation Metrics are Optimal Transport Distances, and Can be Computed Efficiently (NeurIPS 2024 · 1 citation) | |
| use the Wasserstein gradient rather than the L2 gradient to update the barycenter | simulation studies · performance | Helped | Optimal Transport Barycenter via Nonconvex-Concave Minimax Optimization (ICML 2025 · 0 citations) | |
| use an adaptive temperature schedule instead of fixed q=2^{1/2} | Upsampled MNIST (n=4096), L1 distance · seconds | 1.56 -> 1.55 | No clear effect | A Truncated Newton Method for Optimal Transport (ICLR 2025 · 0 citations) |
| Set ε = 0 instead of ε = 1 | CIFAR10, ResNet20, sparsity 0.95, 10% noisy data, noise = σ · testing loss | 2.78 -> 2.77 | Helped | SWAP: Sparse Entropic Wasserstein Regression for Robust Network Pruning (ICLR 2024 · 0 citations) |
| add the low-rank term to the sparse approximate Hessian | Synthetic I and Synthetic II, n = m = 1000, 5000, and 10000 · gradient norm convergence | Helped | The Sparse-Plus-Low-Rank Quasi-Newton Method for Entropic-Regularized Optimal Transport (ICML 2025) |
What reviewers will expect
Datasets and baselines named in the 18 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MNIST · 2 of 18 papers
Baselines
- APDAGD · 2 of 18 papers
- OTCFM · 2 of 18 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 45 papers; from OpenAlex and the papers' affiliations).
- Virginia Polytechnic Institute and State University · 3 papers
- Google DeepMind · 2 papers
- Walmart Global Tech · 2 papers
- Virginia Tech · 2 papers
- North Carolina State University · 2 papers
- Microsoft · 2 papers
- University of British Columbia · 2 papers
- Shanghai University of Finance and Economics · 2 papers
- Peking University · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Neural optimal transport (99 papers)
- Transport map estimation (77 papers)
- Schrodinger bridge matching (11 papers)
Elsewhere in the topic
- Wasserstein barycenters and GW (55 papers)
- Sliced Wasserstein variants (50 papers)
- Wasserstein losses for generative models (23 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).