Mis on voogupõhine aktiivõpe?

Stream-based active learning is an approach in active learning where data arrives to the model in the form of a continuous stream, and the system decides in real-time which new examples should be labeled for training.

Unlike classical pool-based active learning, where there is a fixed set of unlabeled data, stream-based active learning works with sequential data arrival and limited labeling resources.

Main principles:

• The model evaluates each new example as it arrives.
• It selects only the most informative ones (e.g., with high uncertainty) for labeling.
• It allows efficient use of the limited labeling budget under continuous data stream conditions.

Example: in a spam filtering system, new emails arrive constantly, and the model selects only those it doubts for manual review to improve classification quality.