Active-learning label-efficiency study on the VarWISE NEOWISE infrared variable catalog: 86% fewer labels for equal accuracy, with the gain concentrated almost entirely in the rarest variability classes. Includes an independent SIMBAD validation of VarWISE's published classifications.
astronomy xgboost lightgbm class-imbalance active-learning variable-stars astroml label-efficiency varwise
-
Updated
Aug 18, 2026 - Python