Hierarchical boosting for transfer learning with multi-source
Chenglong Wang, Yue Wu, Zongtian Liu · 2016
Transfer learning is a new research direction in the field of machine learning and has achieved good results in classification. However, it is a thorny problem to address the problem of over-fitting and generalization error, and reduce negative transfer, when utilizing the old data to help a small amount of newly labeled data to construct a high-quality classification model. We propose a novel transfer learning method, referred to as Hierarchical Boosting for Transfer Learning with Multi-source (MHTL-AdaBoost), to address over-fitting and generalization error problem, and then reduce negative transfer. Specifically, the unlabeled data in the target are more helpful for transfer learning. MHTL-AdaBoost makes full use of the unlabeled data for clustering and then transfers the source domain data selected to enhance classification. We combine the theories of hierarchical approach and boosting into transfer learning and reduce generalization error. Finally, experimental results show that the proposed algorithm has higher classification accuracy and validate the effectiveness.