How self-supervised learning works: a masked autoencoder finds brain lesions with zero labels and beats training from scratch with few labels
How semi-supervised learning works: FixMatch on brain-MRI-style scans, why it fails on subtle lesions, and how to fix it
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.
PyTorch pipeline for few-shot semantic segmentation of 3D seismic volumes, with configurable data partitioning and support selection.
Label-efficiency benchmark — margin sampling hits target accuracy with 38% fewer labels than random, while least-confidence/entropy underperform it. The honest 'which strategy, not whether'
To associate your repository with the label-efficiency topic, visit your repo's landing page and select "manage topics."