Abstract: Co-training is a popular disagreement-based semi-supervised learning method. Learners of different views mutually select reliable unlabeled instances to augment the labeled dataset. Existing ...