A Survey on Active Deep Learning: From Model Driven to Data Driven

Which samples should be labelled in a large dataset is one of the most important problems for training deep learning. So far, variety active sample selection strategies related to learning have been proposed literature. We defined them as Active Deep Learning (ADL) only if their predictor or selector model, where basic learner called and labeling schemes are selector. In this survey, we categorize ADL into model-driven data-driven by whether its model driven data driven. also introduce different characteristics two major types ADL, respectively. summarized three fundamental factors designation pointed out that, with development learning, experiencing stage from The advantages disadvantages between thoroughly analyzed. Furthermore, sub-classes data-drive discussed emphatically. Finally, survey trend