AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Diagnosing and Improving Diffusion Models by Estimating Optimal Loss Value

Yixian Xu, Shengjie Luo, Liwei Wang, Di He, Chang Liu

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Diffusion theory and design space

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
replace EDM2 schedule with optimal-loss-based scheduleEDM · CIFAR-10 conditionalFID: 1.94 -> 1.75Improved
optimal-loss-based training scheduleEDM · CIFAR-10 unconditionalFID: 1.98 -> 1.94Improved
optimal-loss-based training scheduleEDM · ImageNet-64 conditionalFID: 2.44 -> 2.25Improved
optimal-loss-based training scheduleFlow Matching (EDM sampler) · CIFAR-10 conditionalFID: 2.07 -> 1.77Improved
optimal-loss-based training scheduleFlow Matching (EDM sampler) · ImageNet-64 conditionalFID: 3.06 -> 2.29Improved
optimal-loss-based training scheduleLightningDiT · ImageNet-256 w/o CFGFID: 2.29 -> 2.08Improved
optimal-loss-based training scheduleLightningDiT · ImageNet-256 w/ CFGFID: 1.42 -> 1.30Improved
subtract optimal loss from training loss in scaling-law fit (ablation)EDM2 120M-1.5B · ImageNet-64 large noise scalecorrelation coefficient rho: 0.82 -> 0.94Improved
use uncorrected DOL (C=1) instead of corrected cDOL estimator (ablation)optimal-loss estimator · CIFAR-10estimation errorHurt
use epsilon-prediction target with large-sigma fitNCSN (VE-eps) vs EDM · CIFAR-10FID / loss gapHurt

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