Research map / Learning theory and algorithms / Optimal transport
Large-scale optimal transport map estimation using projection pursuit
Cheng Meng, Yuan Ke, Jingyi Zhang, Mengrui Zhang, Wenxuan Zhong, Ping Ma
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 |
|---|---|---|---|
| replace random projection directions with SAVE-selected informative directions | projection-based OTM estimation · Gaussian OTM, d=10 | mean convergence time (sec): 4.8 -> 0.6 | Improved |
| replace random projection directions with SAVE-selected informative directions | projection-based OTM estimation · Gaussian OTM, d=20 | mean convergence time (sec): 24.4 -> 2.1 | Improved |
| replace random projection with PPMM in high dimension | projection-based OTM estimation · Gaussian OTM, d=50 | convergence: RANDOM fails to converge -> 5.5 | Improved |
| replace sliced method (10 slices) with PPMM | SLICED(10) · Gaussian OTM, d=20 | mean convergence time (sec): 230.2 -> 2.1 | Improved |
| use PPMM instead of linear-program/auction OTM solvers | exact OTM solvers · Gaussian OTM, d=50 | mean convergence time (sec): 46.5 -> 5.5 | Improved |
| increase number of slices in the sliced method | SLICED · Gaussian OTM, d=10/20/50 | estimation accuracy | No clear effect |
| increase slices from 10 to 50 | SLICED · Gaussian OTM, d=10 | CPU time per iteration (sec): 0.111 -> 0.529 | Hurt |
| add SAVE direction selection to random projection | RANDOM · Gaussian OTM, d=50 | CPU time per iteration (sec): 0.015 -> 0.059 | Hurt |
| replace random projection with PPMM for latent-space OT in VAE generation | RANDOM · MNIST | FID: 4.62 -> 0.17 | Improved |
| replace sliced method with PPMM for latent-space OT in VAE generation | SLICED(10) · Doodle (face) | FID: 5.69 -> 0.59 | Improved |
| replace random projection with PPMM | RANDOM · Doodle (cat) | FID: 8.93 -> 0.24 | Improved |
| use SIR instead of SAVE for direction selection (ablation) | PPMM · direction selection | performance | Hurt |
Closest papers by meaning
- Estimation of Stochastic Optimal Transport Maps (NeurIPS 2025 · similarity 0.84)
- On the Private Estimation of Smooth Transport Maps (ICML 2025 · similarity 0.82)
- Stochastic Optimization in Semi-Discrete Optimal Transport: Convergence Analysis and Minimax Rate (NeurIPS 2025 · similarity 0.82)
- Safe and Sparse Newton Method for Entropic-Regularized Optimal Transport (NeurIPS 2024 · similarity 0.82)
- Improving Neural Optimal Transport via Displacement Interpolation (ICLR 2025 · similarity 0.82)
- Geometry-Aware Optimal Transport: Fast Intrinsic Dimension and Wasserstein Distance Estimation (NeurIPS 2026 · similarity 0.81)
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.
