Deep Active Learning: An efficient training methodology with unlabeled data

Although recent advances in the use of deep neural networks for the most different tasks are evident, there is an imitator for the use of this type of network in various problems: the need for a large volume of annotated data. Deep Active Learning is an efficient methodology for acquiring data to be annotated, optimizing the learning process of the model. It proposes the inclusion of the annotator in the training cycle, in order to request its intervention only on data that has a greater potential performance gain.