点击次数：

论文类型：期刊论文

发表时间：2014-02-01

发表刊物：International Journal of Innovative Computing, Information and Control

收录刊物：EI、Scopus

卷号：10

期号：1

页面范围：303-313

ISSN号：13494198

摘要：Transfer learning solves the problem that the training data from a source domain and the test data from a target domain follow different distributions. In this paper, we take advantage of existing well labeled data and introduce them as sources into a novel transductive transfer learning framework. We first construct two feature mapping functions based on mutual information to re-weight the training and the test data. Then we compute the KL-divergence between the posterior probability of the unlabeled data and the prior probability of the labeled data to assign a pseudo-label to the unlabeled data. Next, a set of high-confidence newly-labeled data besides the labeled data are used for training a new classifier. The proposed, algorithm requires that all unlabeled data in the target domain are available during training which is similar to the transductive learning setting, so we call it transductive transfer learning. The effectiveness of the proposed algorithm to transfer learning is verified by experiments in sentiment classification. ? 2014.