AI Paper Atlas

Research map / Learning theory and algorithms / Optimal transport

Importance Sparsification for Sinkhorn Algorithm

Mengyu Li, Jun Yu, Tao Li, Cheng Meng

JMLR 2023 · Learning theory and algorithms › Optimal transport · approach: Sinkhorn and entropic OT algorithms · 8 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
use Spar-Sink instead of Rand-Sink at s=s0(n)Sinkhorn subsampling · Echocardiogram ED prediction, original scaleED prediction error: 0.21 -> 0.09Improved
use Spar-Sink instead of Nys-Sink at s=8s0(n)Sinkhorn subsampling · Echocardiogram ED prediction, original scaleED prediction error: 0.32 -> 0.06Improved
use Spar-Sink instead of full SinkhornWFR distance computation · Echocardiogram ED prediction, original scaleCPU time (s): 15649.31 -> 357.46Improved
increase subsample size from s0(n) to 8s0(n) for Spar-Sink (ablation)Spar-Sink · Echocardiogram ED prediction, original scaleED prediction error: 0.09 -> 0.06Improved
mean-pool images from 112x112 to 56x56 before SinkhornSinkhorn · Echocardiogram ED predictionED prediction error: 0.06 -> 0.11Hurt
mean-pool images before SinkhornSinkhorn · Echocardiogram ED predictionCPU time (s): 15649.31 -> 668.81Improved
use Spar-Sink on original images instead of Sinkhorn on pooled imagesWFR distance computation · Echocardiogram ED predictionED prediction error: 0.11 -> 0.06Improved
use Spar-Sink on pooled images instead of Rand-Sink at s=s0(n)Sinkhorn subsampling · Echocardiogram ED prediction, pooledED prediction error: 0.38 -> 0.30Improved
use sparsified Sinkhorn in the autoencoder (SSAE) instead of Sinkhorn (SAE)Sinkhorn autoencoder · Image generator learning (digits)FID: 24.72 -> 23.65Improved
use SSAE instead of SAESinkhorn autoencoder · Image generator learning (digits)time per epoch (s): 125.87 -> 64.81Improved
decrease epsilon in Sinkhorn (ablation)classical Sinkhorn · Synthetic OT, CPU timeCPU timeHurt
use Spar-Sink instead of Rand-Sink and Nys-Sink on synthetic OT and UOTSinkhorn subsampling · Synthetic data C1-C3RMAEImproved

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