Online Bayesian Imbalanced Learning with Bregman-Calibrated Deep Networks

Zahir Alsulaimawi · 2026

Read Online Bayesian Imbalanced Learning with Bregman-Calibrated Deep Networks by Zahir Alsulaimawi (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-2602-08128.txt.

Description

Classifiers trained on imbalanced data fail minority classes, and when class proportions shift after deployment, most fixes demand fresh training or fresh labeled data. Zahir Alsulaimawi shows online Bayesian imbalanced learning, a method that adapts to shifting priors in real time without fresh training.

The key step separates likelihood-ratio estimation from class-prior assumptions. Building on the link between Bregman divergences and proper scoring rules, the author shows deep networks trained with such losses yield posterior estimates. From those estimates, prior-invariant likelihood ratios can be extracted and proven valid under shift. Fraud detection, medical diagnosis, and anomaly detection, where rare classes matter and distributions move, are the natural beneficiaries.

Catalog record

CREATORZahir Alsulaimawi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE84 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-08128

Browse the full library · About the Tower of Babel