Machine learning this, machine learning that! You know the drill. Let’s talk about a topic that people are only whispering about at the moment, Active Learning.

Active learning is a sub-field of Artificial Intelligence which is based on the fact that curious algorithms are better learners both in terms of efficiency and **expressivity. **The core idea is to let the algorithm pick examples to be trained on rather than the model be trained on all available training data.

Active Learning Scenarios

Active Learning is perhaps one of the simplest ideas in the field of AI. There are multiple variations to this idea but all of them have the stated theme.

Let the model pick the training data

The statement “Let the model pick the training data” can mean several things.

1. Let the model create training data (Generative Models)
2. Let the model pick an example from a stream of unlabelled data
3. Let the model pick an example from a pool of unlabelled data

Generative models can be somewhat tricky because human annotators can have a hard time labeling the data which defeats the purpose of Active Learning.

Generally, in a practical setting, the model picks examples from a stream/pool of unlabelled data to be tagged by human annotators.

#information-theory #python #machine-learning #active-learning #classification