Dynamic Batch Mode Active Learning via L1 Regularization

Active learning algorithms strategy to simultaneously decide the batch size as well as automatically select the exemplar data instances from identify the informative points to be selected for manual annotation, an unlabeled set and thereby reduce human annotation effort through a single framework. Our method has the in training a classifier. Conventional methods of active same complexity as the state-of-the-art static BMAL technique, learning have focused on the pool-based strategy where the where the batch size is pre-specified by the user.