StarCLR: Contrastive Learning Representation for Astronomical Light Curves Astropy ( 3 ; 2 ) , Pandas ( 31 ) , NumPy ( 22 ) , Matplotlib ( 24 ) , PyTorch ( 1 ) , Transformers ( 46 ) , Accelerate ( 21 ) , Datasets ( 27 )

Junyao Ding, Xiaodian Chen, Xinyi Gao, Xiaoyu Tang, Shu Wang, Yang Huang, Xinyu Qi, Guirong Xue, Ali Luo, Jifeng Liu · 2026

Read StarCLR: Contrastive Learning Representation for Astronomical Light Curves Astropy ( 3 ; 2 ) , Pandas ( 31 ) , NumPy ( 22 ) , Matplotlib ( 24 ) , PyTorch ( 1 ) , Transformers ( 46 ) , Accelerate ( 21 ) , Datasets ( 27 ) by Junyao Ding, Xiaodian Chen, Xinyi Gao, Xiaoyu Tang, Shu Wang, Yang Huang, Xinyu Qi, Guirong Xue, Ali Luo, Jifeng Liu (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-2604-24516.txt.

Description

Ding and nine colleagues build StarCLR, a pretraining scheme that teaches a model the shape of star brightness curves by matching overlapping slices of the same curve. Trained on TESS data and tested on three surveys, it sorts variable stars with macro scores of about 84, 88, and 93 percent. It beats from-scratch rivals on TESS and ZTF, most on sparse ZTF curves. Gains shrink on the widest survey, where star physics matters more than the pretrained backbone.

Catalog record

CREATORJunyao Ding, Xiaodian Chen, Xinyi Gao, Xiaoyu Tang, Shu Wang, Yang Huang, Xinyu Qi, Guirong Xue, Ali Luo, Jifeng Liu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE86100
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.24516

Browse the full library · About the Tower of Babel