In Depth
Active learning is an iterative approach where the model identifies which unlabeled data points would be most informative if labeled, and requests human annotation for those specific examples. Instead of randomly labeling data, the model focuses human effort on the examples where it is most uncertain or where labeling would most improve performance.
Common strategies for selecting examples include uncertainty sampling (choosing examples the model is least confident about), query-by-committee (choosing examples where multiple models disagree), and expected model change (choosing examples that would most change the model's parameters). These strategies can reduce the amount of labeled data needed by 10-100x compared to random selection.
Active learning is especially valuable in domains where labeling is expensive: medical diagnosis (requires radiologists), legal document review (requires lawyers), or specialized manufacturing inspection (requires domain experts). It creates a human-in-the-loop workflow where AI and humans collaborate efficiently, with the AI directing human attention to where it matters most.