Définition
Active learning helps you save time and budget on data labeling. Instead of labeling thousands of random examples, the system finds the cases in which the model doubts the most. The expert marks them, and the model quickly improves the quality.
Exemple
For an image moderation system, a model can ask a person to mark controversial images where she is not sure whether this is acceptable content or a violation.
Pourquoi c'est important
This is important for projects where markup is expensive: medicine, legal documents, industrial control, rare languages or complex moderation.
Fonctionnement
The model is trained on a small sample, evaluates unlabeled data, selects the most vague or informative examples, receives expert responses, and repeats the cycle.
Où c'est utilisé
- data markup
- training models with an expert
- improving quality on a small budget
Limites
The method does not help if the initial data is bad or the expert labels inconsistently. It is important to ensure that the model does not become fixated on a narrow group of complex examples.
FAQ
Why is “Active Learning” useful to know?
This is important for projects where markup is expensive: medicine, legal documents, industrial control, rare languages or complex moderation.