Research map / Learning theory and algorithms / Optimal transport
Score-based Generative Neural Networks for Large-Scale Optimal Transport
Max Daniels, Tyler Maunu, PAul HAnd
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 |
|---|---|---|---|
| sample the OT plan with SCONES instead of barycentric projection | entropy-regularized OT · CelebA Super-res., KL lambda=0.1 | FID: 193.92 -> 35.59 | Improved |
| sample the OT plan with SCONES instead of barycentric projection | entropy-regularized OT · CelebA Identity, KL lambda=0.1 | FID: 195.64 -> 36.62 | Improved |
| SCONES instead of barycentric projection with chi2 regularization | chi2-regularized OT · CelebA Super-res., chi2 lambda=0.001 | FID: 212.72 -> 25.59 | Improved |
| SCONES instead of barycentric projection with chi2 regularization | chi2-regularized OT · CelebA Identity, chi2 lambda=0.001 | FID: 214.90 -> 27.88 | Improved |
| lower KL regularization from lambda=0.01 to 0.005 | SCONES · CelebA Super-res. | FID: 35.77 -> 43.80 | Hurt |
| SCONES versus unregularized W2 GAN | OT sampling · CelebA Super-res. | FID: 55.77 -> 35.59 | Improved |
| SCONES versus unregularized W2 Generative Network | OT sampling · CelebA Identity | FID: 20.57 -> 36.62 | Hurt |
| SCONES versus ideally tuned unconditional score network | CelebA generation · CelebA | FID: 10.23 -> 25.59 | Hurt |
| SCONES instead of barycentric projection in 2 dimensions | KL-regularized Gaussian OT · Gaussians d=2 | BW-UVP: 7.1 -> 0.025 | Improved |
| SCONES instead of barycentric projection in 256 dimensions | KL-regularized Gaussian OT · Gaussians d=256 | BW-UVP: 41 -> 2.0 | Improved |
| increase Gaussian dimension from 2 to 256 | SCONES · Gaussian OT | BW-UVP: 0.025 -> 2.0 | Hurt |
| parametrize dual variables with convolutional networks instead of FCNs | Algorithm 1 under L2 cost · OT dual training | optimization quality | No clear effect |
Closest papers by meaning
- On Fitting Flow Models with Large Sinkhorn Couplings (NeurIPS 2026 · similarity 0.88)
- Analyzing and Improving Optimal-Transport-based Adversarial Networks (ICLR 2024 · similarity 0.87)
- Accelerating Sinkhorn algorithm with sparse Newton iterations (ICLR 2024 · similarity 0.87)
- Low-Rank Optimal Transport through Factor Relaxation with Latent Coupling (NeurIPS 2024 · similarity 0.86)
- Hierarchical Refinement: Optimal Transport to Infinity and Beyond (ICML 2025 · similarity 0.86)
- Expectile Regularization for Fast and Accurate Training of Neural Optimal Transport (NeurIPS 2024 · similarity 0.86)
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.
