Active Learning
Let the model pick which examples it’s most confused by, and label only those.
Think of it like
A student who studies the questions they keep getting wrong instead of re-reading the whole textbook.
Example
A vision model flags the 200 images it’s least sure about; a human labels just those and accuracy jumps.
How it actually works
Labeling is expensive, so active learning spends that budget where it helps most: on the examples the model is most uncertain about or that would most reduce its errors. You train, ask the model where it’s weak, label those, and repeat. Done well it reaches the same accuracy with a fraction of the labels.
For product teams
Cuts labeling cost by focusing human effort on the examples that actually move the needle.
For engineers
An iterative loop that selects high-uncertainty or high-information samples for labeling to maximize accuracy per label.
Related
- Gold Labels — Feeds a smaller, smarter gold set.
- Data Flywheel — Powers the loop in a data flywheel.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome