Pseudo-label self-training model for transfer learning algorithm

Abstract When aligning joint distributions between domains, the existing transfer learning algorithms usually assign pseudo labels due to lack of in target domain. However, noise will affect performance learning. Pseudo-label self-training for (PST-TL) model is proposed generate reliable domain and have a wide range applications algorithms. Pseudo are predicted by an ensemble classifier using absolute majority vote, successfully considered be high confidence. The training applies strategy, adding strongly stable data set classifier. semi-supervised unsupervised tasks experiment show that algorithm can significantly improve after embedded PST-TL model.