Fractional Diffusion Bridge Models

Gabriel NobisFraunhofer HHI, Maximilian SpringenbergFraunhofer HHIArina BelovaFraunhofer HHIRembert DaemsGhent University--imecFlandersMake--MIROChristoph KnochenhauerTechnical University of MunichManfred OpperTechnical University of BerlinUniversity of PotsdamUniversity of Birmingham, Tolga BirdalI · 2025

Read Fractional Diffusion Bridge Models by Gabriel NobisFraunhofer HHI, Maximilian SpringenbergFraunhofer HHIArina BelovaFraunhofer HHIRembert DaemsGhent University--imecFlandersMake--MIROChristoph KnochenhauerTechnical University of MunichManfred OpperTechnical University of BerlinUniversity of PotsdamUniversity of Birmingham, Tolga BirdalI (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01795.txt.

Description

Gabriel Nobis and seven colleagues build generative models with memory. Standard diffusion models are driven by Brownian motion, which has no memory: each step forgets the last. Real processes, from protein motion to natural imagery, show long-range correlations, roughness, and anomalous diffusion that Brownian drivers cannot reproduce.

Fractional Diffusion Bridge Models swap in a Markovian approximation of fractional Brownian motion, keeping the process tractable while preserving its non-Markovian character. The authors prove a coupling-preserving bridge exists in this setting. They apply it to predicting future states from paired data. They also extend the setup to the Schrödinger bridge problem for unpaired translation, with a principled loss. On protein conformation prediction it lowered positional error, and on unpaired image translation it beat Brownian baselines on image quality scores.

Catalog record

CREATORGabriel NobisFraunhofer HHI, Maximilian SpringenbergFraunhofer HHIArina BelovaFraunhofer HHIRembert DaemsGhent University--imecFlandersMake--MIROChristoph KnochenhauerTechnical University of MunichManfred OpperTechnical University of BerlinUniversity of PotsdamUniversity of Birmingham, Tolga BirdalI
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE137 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01795

Browse the full library · About the Tower of Babel