AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

A Random Matrix Perspective on the Consistency of Diffusion Models

Binxu Wang, Jacob A Zavatone-Veth, Cengiz Pehlevan

ICML 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
use the linear Gaussian (Wiener filter) predictor to predict generated samplesdiffusion models trained on disjoint splits · FFHQ32 cross-split MSEper-sample Pearson r with cross-split consistency: r=0.244Improved
train on splits stratified along a principal component so mean/variance mismatch (ablation)UNet and DiT diffusion models · FFHQ32 counterfactual splitscross-split sample consistencyHurt
increase training set size n from the memorization regime to n >= 3000 (ablation)UNet/DiT diffusion · FFHQ64 and othersgeneralization and cross-split consistencyImproved
limited training data (finite n) in linear denoiser (ablation)linear denoiser · FFHQ spectrumvariance along low-variance modesHurt
increase dataset size on top versus lower eigenmodes (ablation)UNet/DiT · cross-split MSE per eigenmodecross-split MSEMixed
use DiT instead of UNetdiffusion denoiser · FFHQ64cross-split MSEImproved
train much longer (250k vs 50k steps) (ablation)FFHQ64 UNet · FFHQ64closeness to linear predictor and cross-split consistencyHurt
amplify or shrink initial noise in top-50/200 eigenspace (ablation)UNet FFHQ64 sampling · FFHQ64visual sample qualityMixed
manipulate initial noise in bottom-200 eigenspace (ablation)UNet FFHQ64 sampling · FFHQ64visual sample changeNo 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