AI Paper Atlas

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

NeurIPS 2022 · Learning theory and algorithms › Optimal transport · approach: Transport map estimation · 1 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
select SSNB smoothness parameters by minimizing the semi-dual instead of oracle choiceSSNB · Synthetic Tenserror e_mu: 0.006 -> 0.006No clear effect
select Sinkhorn epsilon by semi-dual vs oracle epsilonSinkhorn · Synthetic Quaderror e_mu: 0.036 -> 0.047Mixed
select Sinkhorn epsilon by semi-dual vs oracle epsilonSinkhorn · Synthetic Tenserror e_mu: 0.059 -> 0.119Hurt
select ICNN parameters by semi-dual vs oracleICNN · Synthetic Quaderror e_mu: 5.11 -> 12.23Hurt
select ICNN parameters by semi-dual vs oracleICNN · Synthetic Tenserror e_mu: 2.74 -> 2.74No clear effect
select among three tuned models with the semi-dualICNN/Sinkhorn/SSNB · Synthetic Quaderror e_mu: 0.013 -> 0.014No clear effect
increase Sinkhorn sample size from 1024 to 10000Sinkhorn tuned with semi-dual · Synthetic Quaderror e_mu: 0.047 -> 0.0104Improved
remove quadratic strong-convexity regularization of ICNN and Sinkhorn potentials (ablation)semi-dual computation · Synthetic OTLegendre transform stabilityHurt
use semi-dual-selected potential instead of best-accuracy potentialICNN · Caltech-office DA A/W1-NN accuracy: 0.36 -> 0.07Hurt
use semi-dual-selected potential instead of best-accuracy potentialSSNB · Caltech-office DA A/W1-NN accuracy: 0.79 -> 0.71Hurt
use semi-dual-selected potential instead of best-accuracy potentialSinkhorn · Caltech-office DA C/A1-NN accuracy: 0.91 -> 0.82Hurt
use semi-dual-selected potential instead of best-accuracy potentialSSNB · Caltech-office DA C/W1-NN accuracy: 0.83 -> 0.83No 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