Active learning

With active learning, humans and a machine model can collaborate to annotate your dataset quicker.

This means that if you start a project from scratch:

1. Humans start annotating manually.
2. Human-generated labels are fed to a machine learning model.
3. After learning on human-generated labels, the machine learning model starts to make predictions.
4. Humans no longer have to annotate. They just review pre-annotated labels generated my a machine learning model.

Thanks to active learning, you can expect a reduction in the number of samples to label to achieve the same performance by up to 50%. This number will depend on the dataset and the task, of course.

In a demo use case with medical image classification (learn more on Kili Active Learning), we experienced an increase from 78% to 85% in accuracy with the same number of samples, and a 30% reduction in the number of samples needed to reach 77% accuracy.