A strategy may prioritize uncertainty, diversity, disagreement, or estimated error, then send those examples for annotation. The model is retrained in rounds, and selection bias must be considered during evaluation.
Active learning selects the unlabeled examples whose labels are expected to be most useful for improving a model.
A strategy may prioritize uncertainty, diversity, disagreement, or estimated error, then send those examples for annotation. The model is retrained in rounds, and selection bias must be considered during evaluation.