Semi-supervised Learning with Transfer Learning

Abstract

Traditional machine learning works well under the assumption that the training data and test data are in the same distribution. However, in many real-world applications, this assumption does not hold. The research of knowledge transfer has received considerable interest recently in Natural Language Processing to improve the domain adaptation of machine learning. In this paper, we present a novel transfer learning framework called TPTSVM (Transfer Progressive Transductive Support Vector Machine), which combines transfer learning and semi-supervised learning. TPTSVM makes use of the limited labeled data in target domain to leverage a large amount of labeled data in source domain and queries the most confident instances in target domain. Experiments on two data sets show that TPTSVM algorithm always improves the classification performance compared to other state-of-the-art transfer learning approaches or semi-supervised approaches. Furthermore, our algorithm could be extended to multiple source domains easily.