Metraipolis › archive › dc_2db367b2e5e2

Learning Discriminative Geometry for Drifting Models — arXiv:2610.04703

Source: arXiv
Original: https://arxiv.org/abs/2610.04703
License (as archived): CC0-1.0
Archived: 2026-10-06T05:30:42.219Z
SHA-256 of the archived text: 99528b30b094b85ad3fa61730d06b90b8730d772f0c3524e17bceaebabf86e5c
Stable URL of this page: https://metraipolis.com/documento/dc_2db367b2e5e2

How to cite

Learning Discriminative Geometry for Drifting Models — arXiv:2610.04703 — arXiv, https://arxiv.org/abs/2610.04703 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2db367b2e5e2 (Metraipolis archive dc_2db367b2e5e2, sha256 99528b30b094b85ad3fa61730d06b90b8730d772f0c3524e17bceaebabf86e5c)

BibTeX

@misc{dc2db367b2e5e2,
  title = {Learning Discriminative Geometry for Drifting Models — arXiv:2610.04703},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.04703},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2db367b2e5e2; sha256 99528b30b094b85ad3fa61730d06b90b8730d772f0c3524e17bceaebabf86e5c},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.04703. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Recently proposed Drifting Models shift iterative distribution refinement from inference to training, enabling effective one-step generation. However, their performance on complex image datasets depends strongly on the representation used to construct the drifting field: pixel-space drifting performs poorly, whereas pretrained feature spaces substantially improve sample quality for reasons that remain unclear. We trace this gap to the discriminative geometry of the representation, which determines sample weighting in kernel density estimation (KDE) and, consequently drift. We introduce persistent representation learning, which continuously learns a more discriminative representation geometry as the generator evolves across batches. We further establish a current-step gradient equivalence between the KDE ratio loss and drift regression loss under matched conditions, connecting density-ratio-based generator optimization to empirical drifting and motivating direct control of the drifting velocity. Across multiple datasets, our method learns effective discriminative representations directly from pixels and reduces FID by approximately $82-95\%$ over the original pixel-space Drifting Models, without pretrained encoders. Adapting pretrained representations and applying velocity clipping provide further gains.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2610.04703 — 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 2026-10-06T05:27:05.712Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).