AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

MAD: Manifold Attracted Diffusion

Dennis Elbrächter, Giovanni S Alberti, Matteo Santacesaria

ICML 2026 · Image generation › Fast diffusion sampling · approach: Diffusion theory and design space · 0 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
replace standard score inference with extended score inference (MAD)DDPM++ trained on noisy synthetic shapes · synthetic data, Gaussian noiseFID: 319.42 -> 189.51Improved
replace standard score inference with MADDDPM++ trained on noisy CIFAR-10 · CIFAR-10, Gaussian noiseFID: 147.47 -> 77.15Improved
replace standard score inference with MADDDPM++ trained on noisy FFHQ · FFHQ, Gaussian noiseFID: 134.42 -> 74.52Improved
replace standard score inference with MAD on non-Gaussian corruptionDDPM++ trained on FFHQ with pixel removal · FFHQ, pixel removalFID: 387.87 -> 104.00Improved
replace standard score inference with MAD on shot noiseDDPM++ trained on FFHQ with shot noise · FFHQ, shot noiseFID: 257.42 -> 101.12Improved
replace standard score inference with MAD on pixel removal for synthetic dataDDPM++ trained on synthetic shapes · synthetic data, pixel removalFID: 234.40 -> 163.14Improved
train diffusion model on noisy data and use standard score inferenceDDPM++ · all noisy datasetsFID vs clean data: 320.56 -> 319.42 (synthetic Gaussian training data vs standard generation)No clear effect
increase parameter b in extended score inference (ablation)MAD on pretrained FFHQ and AFHQv2 networks · noiseless FFHQ/AFHQv2background and fine featuresMixed
use extended score only in the last half of inference steps (ablation)MAD · inference costcompute overhead: 100% -> 50% additional costImproved
compute exact derivative of score network via automatic differentiation instead of finite difference (ablation)MAD · numerical experimentssample quality and costNo clear effect
apply MAD to Cryo-EM EMPIAR-11618 with a network trained only on noisy dataDDPM++ · EMPIAR-11618qualitative particle shapesImproved

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