Mis on basseini põhine aktiivõpe?

Pool-based active learning is an approach in active learning where the model selects the most informative examples from a large pool of unlabeled data.

Main idea:

• There is a large set of unlabeled data (pool).
• The model is trained on the current labeled set.
• It then evaluates the unlabeled data and selects those that, in its opinion, will provide the greatest benefit when labeled (e.g., with the highest uncertainty).
• These selected examples are sent for labeling (e.g., to an expert).
• After labeling, the data are added to the training set, and the cycle repeats.

This approach allows for efficient use of labeling resources, focusing on the most useful data, which is especially important in cases of weak supervision and limited resources.