Architecture

L1 training data

L1 learner

L1 model

training data

L2 training data

L2 learner

L2 model

unlabeled data

• B&M co-training handles only binary classification • Handles distribution skewing