Practice · Established · Advanced

Active Learning

Letting the model choose which examples to have labelled next, so annotation effort goes where it changes the model most.

What Active Learning is

Instead of labelling randomly, active learning prioritises the examples the model is least certain about or that best represent unlabelled clusters.

How it works

Uncertainty sampling, margin sampling and diversity-based selection rank candidates; a human labels the top batch; the model retrains and the loop repeats.

Why it matters

It often reaches target accuracy with a fraction of the labels, which matters when annotation requires scarce expert time.

Common uses

→Medical and legal annotation
→Rare-class detection
→Bootstrapping moderation systems

Strengths

✓Large labelling cost savings

Watch for

✓Selection bias in the resulting dataset
✓Needs retraining infrastructure

Continue exploring

Data Labeling Semi-Supervised Learning Human in the Loop