Label transfer and propagation for domain adaptation

Traditional classification algorithms often perform well when training and testing data are drawn from the identical distribution. However, in real applications, this condition may be not satisfied. Domain adaptation is an effective approach to deal with this problem. In this paper, we propose an efficient two-stage algorithm for domain adaptation. In the label transfer stage, we utilize training classifier to predict testing data with different weights (confidence) based on their signed distance to the domain separator, which is a classifier maximally separating training data (from source domain) and testing data (from target domain) apart. In the label propagation stage, we introduce manifold regularization to propagate the labels of target data with larger weights to ones with smaller weights. Furthermore, the target classifier can be obtained in a closed form. The extensive experiments on an artificial dataset and a real benchmark verify the effectiveness of our approach. Empirical results demonstrate that the proposed method is competitive with state-of-the-art domain adaptation algorithms.