in which the model keeps predicting the class of each unlabeled sample and learns from the feedback that whether 2