Active Learning
A machine learning approach where the model identifies which unlabeled examples would be most informative to label next, minimizing annotation effort.
How It Works
The model is trained on a small labeled dataset. It then scores unlabeled examples by uncertainty, diversity, or expected impact. The most informative examples are sent for human labeling, and the cycle repeats.
Benefits
Can achieve target performance with 10-50% fewer labeled examples compared to random sampling. Especially valuable when labeling is expensive (medical imaging, legal documents).
Strategies
Uncertainty sampling: Label examples the model is least confident about. Query by committee: Label examples where multiple models disagree. Expected model change: Label examples that would most change the model.