Denoising growth complexity: Data geometry and certified schedules for diffusion sampling

DOI10.48550/arXiv.2607.26285MaRDI QIDQ7284585arXivFDO

Authors Martin J. Wainwright

Publication date 28 July 2026

Published in arXiv

Tldr We introduce denoising growth complexity ($\mathsf{DGC}$) as a geometric measure of diffusion-based sampling algorithms' performance and derive certified schedules based on it that outperform uniform stepsize in certain regimes.

Full work available at URL https://doi.org/10.48550/arXiv.2607.26285

arXiv classification

cs.LGmath.NAmath.STstat.ML

This page was built for publication: Denoising growth complexity: Data geometry and certified schedules for diffusion sampling

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7284585)