Active learning is a case of machine learning where a model can query a user or some other type of information system to obtain the desired outputs at new data points. In other words, in active learning, the learning algorithm can actively choose which examples it wants to learn from.
The goal of active learning is to minimize the number of labeled instances needed to learn a good classifier, as labeling instances can be expensive or time-consuming. The model tries to select the most informative samples to learn from, rather than using a randomly selected set of labeled samples.
There are several strategies in active learning, such as uncertainty sampling, query-by-committee, and expected model change. These techniques focus on selecting samples for which the model is most uncertain, has the highest disagreement among multiple models, or is expected to change the current model the most, respectively.
Active learning has applications in various fields where labeled data are scarce or expensive to obtain, such as medical imaging, text classification, and more. It can significantly reduce the amount of data needed to train a model, making it a valuable approach in many scenarios.