Metraipolis › archive › dc_7e5bd3753a89

On the Diffusibility of High-Dimensional Latents — arXiv:2609.28473

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.28473
License (as archived)
CC0-1.0
Archived
2026-10-08T10:58:25.045Z
SHA-256 of the archived text
ad6d725dfcb95aa229debb496a0957a6b38f72c5b18ce2ab60eff86a9f7eb970
Stable URL of this page
https://metraipolis.com/documento/dc_7e5bd3753a89

How to cite

On the Diffusibility of High-Dimensional Latents — arXiv:2609.28473 — arXiv (en), https://arxiv.org/abs/2609.28473 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7e5bd3753a89 (Metraipolis archive dc_7e5bd3753a89, sha256 ad6d725dfcb95aa229debb496a0957a6b38f72c5b18ce2ab60eff86a9f7eb970)
BibTeX
@misc{dc7e5bd3753a89,
  title = {On the Diffusibility of High-Dimensional Latents — arXiv:2609.28473},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.28473},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7e5bd3753a89; sha256 ad6d725dfcb95aa229debb496a0957a6b38f72c5b18ce2ab60eff86a9f7eb970},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.28473. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Representation Autoencoders (RAEs) enable diffusion models to operate in the feature spaces of pretrained visual encoders. However, many off-the-shelf encoders are not optimized for faithful reconstruction, discarding fine-grained visual details. As expected, finetuning these encoders for image reconstruction recovers such details. However, perhaps counterintuitively, this procedure reduces the effective dimensionality of the resulting representation, and the altered geometry has downstream effects on generation. Specifically, we show that using the standard velocity prediction in flow matching in this high-dimensional space requires the model to fit orthogonal noise directions outside the low-dimensional signal manifold, making optimization inefficient. This motivates using the clean data parameterization ($\boldsymbol{x}_{0}$-prediction) instead, which focuses learning on the underlying signal manifold. Across experiments with multiple strong-reconstruction encoders, we show that $\boldsymbol{x}_{0}$-prediction consistently improves text-to-image generation performance.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-24 in cs.LG).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.28473 — 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 2026-10-08T10:06:16.853Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-24?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).