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'
python machine-learning annotation scikit-learn ml active-learning uncertainty-sampling sample-selection data-centric-ai label-efficiency
-
Updated
Jun 30, 2026 - Python