Spectral Regularization for Diffusion Models

Satish Chandran, Nícolas Roque dos Santos, Yunshu Wu, Greg Ver Steeg, Evangelos Papalexakis · 2026

Read Spectral Regularization for Diffusion Models by Satish Chandran, Nícolas Roque dos Santos, Yunshu Wu, Greg Ver Steeg, Evangelos Papalexakis (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-02447.txt.

Description

Satish Chandran, Nícolas Roque dos Santos, Yunshu Wu, and colleagues regularize diffusion models in the frequency domain in this 2026 paper. Diffusion models learn to turn noise into images and audio. Standard training scores each point alone, ignoring frequency and scale structure. Natural signals, though, live across many scales at once.

Their losses add Fourier and wavelet terms to the usual training objective. The diffusion process, network design, and sampling all stay unchanged. Those terms act as soft biases toward balanced frequencies and coherent fine detail. The method fits several major diffusion variants with negligible extra cost. Image and audio tests show steady quality gains. High-resolution unconditional tasks, where fine structure is hardest, gained the most.

Catalog record

CREATORSatish Chandran, Nícolas Roque dos Santos, Yunshu Wu, Greg Ver Steeg, Evangelos Papalexakis
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50258
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.02447

Browse the full library · About the Tower of Babel