Active Learning

Labeling data is one of the main bottlenecks in production ML. A single label can cost almost nothing for simple crowdsourced text classification or tens of dollars for expert medical annotation. Active learning helps you spend that labeling budget on examples that are likely to improve the model.

Instead of labeling random examples, the model identifies examples that are uncertain, underrepresented, diverse, or likely to expose a failure mode. This isn't about labeling less for its own sake; it's about spending the budget on the labels with the highest marginal value.