Research map / Learning theory and algorithms / Optimal transport
Meta Optimal Transport
Brandon Amos, Giulia Luise, Samuel Cohen, Ievgen Redko
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 |
|---|---|---|---|
| Initialize Sinkhorn with predicted Meta OT dual potentials instead of zero potentials | Sinkhorn · MNIST | runtime to marginal error 10^{-2}: 4.5\cdot10^{-3} -> 2.3\cdot10^{-3} | Improved |
| Initialize Sinkhorn with predicted Meta OT dual potentials instead of Gaussian initialization | Sinkhorn · Spherical transport | runtime to marginal error 10^{-2}: 0.56 -> 7.8\cdot10^{-2} | Improved |
| Use Meta OT warm starts and 5 Sinkhorn iterations rather than WAR with 20 iterations | WAR classification · Fashion MNIST, 40% label noise | classification accuracy: 88.08 -> 89.08 | Improved |
| Use Meta OT warm starts and 5 Sinkhorn iterations rather than WAR with 20 iterations | WAR classification · Cifar-10, 40% label noise | classification accuracy: 87.07 -> 83.57 | Hurt |
| Use Meta OT warm starts and 5 Sinkhorn iterations rather than WAR with 20 iterations | WAR classification · Cifar-100, 40% label noise | classification accuracy: 61.63 -> 61.13 | No clear effect |
| Replace zero initialization with Meta OT warm starts while using 5 iterations | WAR classification · Fashion MNIST, 0% label noise | classification accuracy: 94.75 -> 94.70 | No clear effect |
| Initialize W2GN from Meta ICNN predictions rather than train from scratch | W2GN color transfer · Color transfer | dual value after 1k iterations: 0.96 -> 1.0 | Improved |
| Predict one discrete dual potential and recover the other using the optimality relation | Meta OT discrete model · MNIST and spherical transport | runtime to convergence | Improved |
| Use an amortized Meta ICNN initialization before W2GN fine-tuning | W2GN color transfer · Color transfer | convergence / dual objective | Improved |
| Use Meta OT warm starts with 5 iterations instead of zero initialization with 20 iterations | WAR training · Fashion MNIST | runtime per epoch: 17.04 -> 14.37 | Improved |
Closest papers by meaning
- Amortized Optimal Transport from Sliced Potentials (NeurIPS 2026 · similarity 0.89)
- Progressive Entropic Optimal Transport Solvers (NeurIPS 2024 · similarity 0.87)
- Expected Sliced Transport Plans (ICLR 2025 · similarity 0.87)
- Improving Neural Optimal Transport via Displacement Interpolation (ICLR 2025 · similarity 0.86)
- Hierarchical Refinement: Optimal Transport to Infinity and Beyond (ICML 2025 · similarity 0.86)
- Safe and Sparse Newton Method for Entropic-Regularized Optimal Transport (NeurIPS 2024 · similarity 0.85)
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.
