Research map / Learning theory and algorithms / Optimal transport
Parameter tuning and model selection in Optimal Transport with semi-dual Brenier formulation
Adrien Vacher, François-Xavier Vialard
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 |
|---|---|---|---|
| select SSNB smoothness parameters by minimizing the semi-dual instead of oracle choice | SSNB · Synthetic Tens | error e_mu: 0.006 -> 0.006 | No clear effect |
| select Sinkhorn epsilon by semi-dual vs oracle epsilon | Sinkhorn · Synthetic Quad | error e_mu: 0.036 -> 0.047 | Mixed |
| select Sinkhorn epsilon by semi-dual vs oracle epsilon | Sinkhorn · Synthetic Tens | error e_mu: 0.059 -> 0.119 | Hurt |
| select ICNN parameters by semi-dual vs oracle | ICNN · Synthetic Quad | error e_mu: 5.11 -> 12.23 | Hurt |
| select ICNN parameters by semi-dual vs oracle | ICNN · Synthetic Tens | error e_mu: 2.74 -> 2.74 | No clear effect |
| select among three tuned models with the semi-dual | ICNN/Sinkhorn/SSNB · Synthetic Quad | error e_mu: 0.013 -> 0.014 | No clear effect |
| increase Sinkhorn sample size from 1024 to 10000 | Sinkhorn tuned with semi-dual · Synthetic Quad | error e_mu: 0.047 -> 0.0104 | Improved |
| remove quadratic strong-convexity regularization of ICNN and Sinkhorn potentials (ablation) | semi-dual computation · Synthetic OT | Legendre transform stability | Hurt |
| use semi-dual-selected potential instead of best-accuracy potential | ICNN · Caltech-office DA A/W | 1-NN accuracy: 0.36 -> 0.07 | Hurt |
| use semi-dual-selected potential instead of best-accuracy potential | SSNB · Caltech-office DA A/W | 1-NN accuracy: 0.79 -> 0.71 | Hurt |
| use semi-dual-selected potential instead of best-accuracy potential | Sinkhorn · Caltech-office DA C/A | 1-NN accuracy: 0.91 -> 0.82 | Hurt |
| use semi-dual-selected potential instead of best-accuracy potential | SSNB · Caltech-office DA C/W | 1-NN accuracy: 0.83 -> 0.83 | No clear effect |
Closest papers by meaning
- Expectile Regularization for Fast and Accurate Training of Neural Optimal Transport (NeurIPS 2024 · similarity 0.88)
- Geometry-Aware Optimal Transport: Fast Intrinsic Dimension and Wasserstein Distance Estimation (NeurIPS 2026 · similarity 0.87)
- Energy-guided Entropic Neural Optimal Transport (ICLR 2024 · similarity 0.87)
- Nonlinear Filtering with Brenier Optimal Transport Maps (ICML 2024 · similarity 0.87)
- Progressive Entropic Optimal Transport Solvers (NeurIPS 2024 · similarity 0.87)
- Improving Neural Optimal Transport via Displacement Interpolation (ICLR 2025 · 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.
