AI Paper Atlas

Research map / Learning theory and algorithms / Optimal transport

Score-based Generative Neural Networks for Large-Scale Optimal Transport

Max Daniels, Tyler Maunu, PAul HAnd

NeurIPS 2021 · Learning theory and algorithms › Optimal transport · approach: Neural optimal transport · 5 citations (OpenAlex)

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 changedWhereMeasuredEffect
sample the OT plan with SCONES instead of barycentric projectionentropy-regularized OT · CelebA Super-res., KL lambda=0.1FID: 193.92 -> 35.59Improved
sample the OT plan with SCONES instead of barycentric projectionentropy-regularized OT · CelebA Identity, KL lambda=0.1FID: 195.64 -> 36.62Improved
SCONES instead of barycentric projection with chi2 regularizationchi2-regularized OT · CelebA Super-res., chi2 lambda=0.001FID: 212.72 -> 25.59Improved
SCONES instead of barycentric projection with chi2 regularizationchi2-regularized OT · CelebA Identity, chi2 lambda=0.001FID: 214.90 -> 27.88Improved
lower KL regularization from lambda=0.01 to 0.005SCONES · CelebA Super-res.FID: 35.77 -> 43.80Hurt
SCONES versus unregularized W2 GANOT sampling · CelebA Super-res.FID: 55.77 -> 35.59Improved
SCONES versus unregularized W2 Generative NetworkOT sampling · CelebA IdentityFID: 20.57 -> 36.62Hurt
SCONES versus ideally tuned unconditional score networkCelebA generation · CelebAFID: 10.23 -> 25.59Hurt
SCONES instead of barycentric projection in 2 dimensionsKL-regularized Gaussian OT · Gaussians d=2BW-UVP: 7.1 -> 0.025Improved
SCONES instead of barycentric projection in 256 dimensionsKL-regularized Gaussian OT · Gaussians d=256BW-UVP: 41 -> 2.0Improved
increase Gaussian dimension from 2 to 256SCONES · Gaussian OTBW-UVP: 0.025 -> 2.0Hurt
parametrize dual variables with convolutional networks instead of FCNsAlgorithm 1 under L2 cost · OT dual trainingoptimization qualityNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy