Dense Feature Learning via Linear Structure Preservation in Medical Data

Yuanyun Zhang, Mingxuan Zhang, Siyuan Li, Zihan Wang, Haoran Chen, Wenbo Zhou, Shi Li · 2026

Read Dense Feature Learning via Linear Structure Preservation in Medical Data by Yuanyun Zhang, Mingxuan Zhang, Siyuan Li, Zihan Wang, Haoran Chen, Wenbo Zhou, Shi Li (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-2602-07706.txt.

Description

Medical AI models often squeeze patient data into a few narrow patterns, which hurts reuse on new tasks. Yuanyun Zhang and six co-authors propose dense feature learning, a training approach that shapes the geometry of medical embeddings directly instead of chasing one prediction target.

The method rewards balanced, well-spread, and mutually distinct features using only linear-algebra rules, with no labels and no attempt to rebuild the input. Tests on hospital records over time, clinical text, and combined patient data show steadier features and better results on later tasks. The paper suits readers building reusable representations for health data rather than one-off predictors.

Catalog record

CREATORYuanyun Zhang, Mingxuan Zhang, Siyuan Li, Zihan Wang, Haoran Chen, Wenbo Zhou, Shi Li
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07706

Browse the full library · About the Tower of Babel