Filtered Posterior Mean Collections: A Unified Framework for Analytical Models of Diffusion Generalization

Matthew Niedoba [email protected], Berend Zwartsenberg [email protected], Frank Wood [email protected], [1em] \AFFS University of British Columbia, Inverted AI, Alberta Machine Intelligence Institute · 2026

Read Filtered Posterior Mean Collections: A Unified Framework for Analytical Models of Diffusion Generalization by Matthew Niedoba [email protected], Berend Zwartsenberg [email protected], Frank Wood [email protected], [1em] \AFFS University of British Columbia, Inverted AI, Alberta Machine Intelligence Institute (2026) 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-2605-24192.txt.

Description

Niedoba, Zwartsenberg, and Wood unify math models of diffusion generalization under Filtered Posterior Mean Collections. Cleaning networks in diffusion models behave alike across designs and settings, and recent work models their outputs by mixing training patches weighted by odds. The authors define the unified class through query sharpness vectors, response weights, and source pools, showing earlier methods emerge from choices. Soft easing of patch methods and grown source pools each improve results, and applied to a known set, the findings yield gains across three natural image data sets.

Catalog record

CREATORMatthew Niedoba [email protected], Berend Zwartsenberg [email protected], Frank Wood [email protected], [1em] \AFFS University of British Columbia, Inverted AI, Alberta Machine Intelligence Institute
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE81279
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2605.24192

Browse the full library · About the Tower of Babel