Metraipolis › archive › dc_8cd9eaea8838
Steepest Guidance: A Practical and Principled Approach to Inference-Time Alignment of Flow and Diffusion-based Models — arXiv:2609.39091
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.39091
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T13:47:28.394Z
- SHA-256 of the archived text
cd4f06327743de452c3239fe55f224f64bce1e2a18f5bcc722869b017837e5e1- Stable URL of this page
https://metraipolis.com/documento/dc_8cd9eaea8838
How to cite
Steepest Guidance: A Practical and Principled Approach to Inference-Time Alignment of Flow and Diffusion-based Models — arXiv:2609.39091 — arXiv (en), https://arxiv.org/abs/2609.39091 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8cd9eaea8838 (Metraipolis archive dc_8cd9eaea8838, sha256 cd4f06327743de452c3239fe55f224f64bce1e2a18f5bcc722869b017837e5e1)
BibTeX
@misc{dc8cd9eaea8838,
title = {Steepest Guidance: A Practical and Principled Approach to Inference-Time Alignment of Flow and Diffusion-based Models — arXiv:2609.39091},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.39091},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8cd9eaea8838; sha256 cd4f06327743de452c3239fe55f224f64bce1e2a18f5bcc722869b017837e5e1},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39091. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Inference-time alignment of flow and diffusion-based models is critical for achieving flexible generative modeling. Theoretically, Doob's $h$-transform provides an elegant solution to this problem, and most existing methods are based on this principle. However, in practice, estimating the optimal guidance derived from Doob's $h$-transform at inference time is challenging. To deal with this issue, we regard inference-time alignment as a sequential optimization problem in the space of probability measures and propose a novel framework called *Steepest Guidance*, based on the principle of maximizing local improvement in the objective. We provide a theoretical analysis of the proposed method and demonstrate its effectiveness through extensive experiments. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.LG). Categorie arXiv: Machine Learning (stat.ML) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.39091 — 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 2026-10-07T13:18:10.617Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).