Exploiting the Asymmetric Uncertainty Structure of Pre-trained VLMs on the Unit Hypersphere

Li Ju, Max Andersson, Stina Fredriksson, Edward Glöckner, Andreas Hellander, Ekta Vats, Prashant Singh · 2025

Read Exploiting the Asymmetric Uncertainty Structure of Pre-trained VLMs on the Unit Hypersphere by Li Ju, Max Andersson, Stina Fredriksson, Edward Glöckner, Andreas Hellander, Ekta Vats, Prashant Singh (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-2505-11029.txt.

Description

Li Ju and co-authors present a 2025 paper on measuring uncertainty in vision-language models. These models turn images and text into points in a shared space, but a single point hides how unsure the match is. Recent fixes attach a probability spread to each point, yet they treat text and image uncertainty as if it had the same shape.

Their method, AsymVLM, builds the spreads directly on the unit sphere where the model points live. It lets the two modalities carry different uncertainty structures, matching how data actually behave. Benchmark tests checked that the spreads improve performance on standard tasks. Follow-up experiments confirmed that the asymmetry is real and not an artifact of the method.

Catalog record

CREATORLi Ju, Max Andersson, Stina Fredriksson, Edward Glöckner, Andreas Hellander, Ekta Vats, Prashant Singh
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-11029

Browse the full library · About the Tower of Babel