Climbing the label tree: Hierarchy-preserving contrastive learning for medical imaging
Alif Elham Khan · 2025
Read Climbing the label tree: Hierarchy-preserving contrastive learning for medical imaging by Alif Elham Khan (2025) 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-2511-03771.txt.
Description
Alif Elham Khan presents a self-supervised learning method for medical images whose labels form a tree, running from organ down to tissue and subtype. Standard training ignores that hierarchy. Khan makes the label tree a direct training signal and also a target for judging how faithful the learned features remain.
Two added objectives do the work. One weights contrastive pairs by shared ancestors, and the other uses a margin to separate groups at different levels. Tests on several benchmarks, including breast histopathology, show better features than strong baselines and closer respect for the taxonomy. The approach works with flat or curved embeddings and needs no change to the model design.
Catalog record
| CREATOR | Alif Elham Khan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-03771