Active learning is when the input of a labeler is asked during training to optimally deal with the model uncertainty. This approach is part of the general semi supervised tasks as it suggest to label iteratively a minimal number of new examples to train a model.