Magnification-Invariant Image Classification via Domain Generalization and Stable Sparse Embedding Signatures

Ifeanyi Ezuma, Olusiji Medaiyese · 2026

Read Magnification-Invariant Image Classification via Domain Generalization and Stable Sparse Embedding Signatures by Ifeanyi Ezuma, Olusiji Medaiyese (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-2604-25817.txt.

Description

Ezuma and Medaiyese tackle magnification shift in histopathology classification, where models trained at one scale fail at another. They test on BreaKHis under a strict patient-disjoint protocol holding out one magnification at a time. Compared are a supervised baseline, baseline plus GAN patches, and a domain-general model using gradient reversal. The domain-general model wins overall, clearest at 200X held out. It also shrinks sparse signatures more than threefold at equal predictive power, with far better cross-fold reproducibility.

Catalog record

CREATORIfeanyi Ezuma, Olusiji Medaiyese
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE27714
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.25817

Browse the full library · About the Tower of Babel