Activation-Space Uncertainty Quantification for Pretrained Networks

Richard Bergna, Stefan Depeweg, Sergio Calvo-Ordoñez, Jonathan Plenk, Alvaro Cartea, Jose Miguel Hernández-Lobato · 2026

Read Activation-Space Uncertainty Quantification for Pretrained Networks by Richard Bergna, Stefan Depeweg, Sergio Calvo-Ordoñez, Jonathan Plenk, Alvaro Cartea, Jose Miguel Hernández-Lobato (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-2602-14934.txt.

Description

Richard Bergna and others give Gaussian Process Activations (GAPA), a late-check doubt gauge for ready-made nets that shifts Bayes work from weights to layer outputs. Strong doubt gauges often ask for re-training, Monte Carlo draws, or dear second-order sums, and may shift a locked core's answers.

GAPA swaps plain bends for Bayes bends whose mean match holds the old bend true, so core answers stay put by build, while closed-form doubt spreads in output space come free. Thin Bayes seed points scale the trick to new nets. Ready-made nets thus gain doubt scores with no re-training and no changed answers.

Catalog record

CREATORRichard Bergna, Stefan Depeweg, Sergio Calvo-Ordoñez, Jonathan Plenk, Alvaro Cartea, Jose Miguel Hernández-Lobato
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE90 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-14934

Browse the full library · About the Tower of Babel