Metraipolis › archive › dc_03d268fcee7c

On the spectral properties of generative denoiser Jacobians — arXiv:2609.36210

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.36210
License (as archived)
CC0-1.0
Archived
2026-10-07T17:28:47.436Z
SHA-256 of the archived text
c050a80ad7db0be53c3f03015010aefae9939f4ebee8b12876514db714d3e6a9
Stable URL of this page
https://metraipolis.com/documento/dc_03d268fcee7c

How to cite

On the spectral properties of generative denoiser Jacobians — arXiv:2609.36210 — arXiv (en), https://arxiv.org/abs/2609.36210 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_03d268fcee7c (Metraipolis archive dc_03d268fcee7c, sha256 c050a80ad7db0be53c3f03015010aefae9939f4ebee8b12876514db714d3e6a9)
BibTeX
@misc{dc03d268fcee7c,
  title = {On the spectral properties of generative denoiser Jacobians — arXiv:2609.36210},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36210},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_03d268fcee7c; sha256 c050a80ad7db0be53c3f03015010aefae9939f4ebee8b12876514db714d3e6a9},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36210. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Generative denoising models, such as diffusion and flow-matching, learn to sample from complex distributions by training a deep neural network denoiser to recover clean data from noise-corrupted samples. While such models are typically compared on the quality of their synthesized samples, these metrics provide limited insight into how the underlying denoiser, which drives generation, differs. In this work, we propose to analyze the spectrum of the denoiser Jacobian as a tool to characterize these differences. Across pre-trained denoising models, we observe that better generative performance is associated with larger Jacobian eigenvalues. Motivated by this, we introduce a regularization scheme that controls the Jacobian spectrum by training the denoiser on perturbed inputs, with perturbations suppressing or amplifying Jacobian responses. On ImageNet, we test whether directly modifying the Jacobian spectral properties leads to improved generations. Our findings suggest that denoisers benefit from both strengthening responses along data-relevant principal eigen-directions and suppressing the noisy, data-irrelevant ones. This establishes the denoiser Jacobian as a useful tool for identifying differences between generative denoising models.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG) ; Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2609.36210 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T16:59:48.117Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).