Reducing Class-Wise Performance Disparity via Margin Regularization

Beier Zhu, Kesen Zhao, Jiequan Cui, Qianru Sun, Yuan Zhou, Xun Yang, Hanwang Zhang · 2026

Read Reducing Class-Wise Performance Disparity via Margin Regularization by Beier Zhu, Kesen Zhao, Jiequan Cui, Qianru Sun, Yuan Zhou, Xun Yang, Hanwang Zhang (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-2602-00205.txt.

Description

Beier Zhu and coauthors reduce class wise accuracy gaps with margin regularization, called MR2. Deep nets often show large gaps across classes even on balanced data, raising doubts about reliable deployment. Their method adjusts margins in logit and representation space, guided by a margin based bound linking per class trait spread to error. Hard classes get larger logit margins, while excess representation margins are penalized to tighten each class. Tests on seven data sets, including ImageNet, show steadier class performance.

Catalog record

CREATORBeier Zhu, Kesen Zhao, Jiequan Cui, Qianru Sun, Yuan Zhou, Xun Yang, Hanwang Zhang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE87656
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00205

Browse the full library · About the Tower of Babel