Metraipolis › archive › dc_a2beda35b0ee
Bidirectional Centered Score Matching for Discrete Diffusion Distillation — arXiv:2611.04279
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.04279
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-04T18:34:31.021Z
- SHA-256 of the archived text
e1ca23fb43a82336e2eec264d67dfb10e8f0299b5f4fe841562122e62ee28749- Stable URL of this page
https://metraipolis.com/documento/dc_a2beda35b0ee
How to cite
Bidirectional Centered Score Matching for Discrete Diffusion Distillation — arXiv:2611.04279 — arXiv (en), https://arxiv.org/abs/2611.04279 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a2beda35b0ee (Metraipolis archive dc_a2beda35b0ee, sha256 e1ca23fb43a82336e2eec264d67dfb10e8f0299b5f4fe841562122e62ee28749)
BibTeX
@misc{dca2beda35b0ee,
title = {Bidirectional Centered Score Matching for Discrete Diffusion Distillation — arXiv:2611.04279},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.04279},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a2beda35b0ee; sha256 e1ca23fb43a82336e2eec264d67dfb10e8f0299b5f4fe841562122e62ee28749},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04279. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Discrete diffusion models with parallel decoding are often distilled using token-wise KL objectives. Although token-wise KL exactly matches the corresponding product-form block distributions, its probability-space optimization geometry and teacher-only candidate truncation may provide insufficient supervision under aggressive few-step distillation. Concrete Score Matching (CSM) provides an alternative within-position objective by aligning relative logit geometry, but its complete pairwise form incurs $\mathcal{O}(|\mathcal{V}|^2)$ complexity. We propose Bidirectional Centered Score Matching (BCSM), an efficient within-position distillation objective that constructs an $\mathcal{O}(K)$ bidirectional active set, reduces pairwise comparisons to $\mathcal{O}(K^2)$, and factorizes them into an $\mathcal{O}(K)$ objective. We further introduce cross-timestep transition alignment. Experiments demonstrate strong quality--efficiency trade-offs, including up to $6\times$ NFE reduction with a 16.9\% relative HumanEval gain and a 15.1\% relative MMMU gain at single-step inference.
Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2611.04279 — 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 2026-11-04T18:32:37.292Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).