Active Learning is a machine learning approach where the model itself identifies which unlabelled examples would be most useful to have labelled, so human effort is focused where it helps the model most.
What it means in plain English
Labelling data is expensive, so it makes sense to label the most informative examples first. In active learning, the model flags the cases it is most uncertain about and requests labels for those, rather than having humans label data at random. By focusing labelling effort on the examples that teach the model the most, it can reach high accuracy with far less labelled data.
It creates a productive loop between the model and human labellers.
A simple example
A model classifying documents labels the easy ones confidently but flags the ambiguous cases it is unsure about. Humans label just those tricky examples, and the model improves quickly — with far less labelling than tagging everything at random.
Why it matters
Active learning makes building models more efficient by reducing how much data must be labelled. In fields where labelling is costly or requires experts — like medicine — it can dramatically cut the effort needed to train an accurate model.
Related terms
- Data Labelling — what active learning makes more efficient.
- Supervised Learning — the setting active learning improves.
- Training Data — built more efficiently with active learning.
Frequently asked questions
How is active learning different from normal supervised learning?
In standard supervised learning you label a fixed dataset up front. In active learning the model itself chooses which unlabeled examples would be most useful to label next, so you get more value from less labelling effort.
When is active learning most useful?
When labelling data is expensive or slow (for example, requiring expert annotators), active learning helps focus that effort on the examples that most improve the model.
