Distribution Adaptation and Discriminative Feature Learning for Unsupervised Domain Adaptation

Unsupervised domain adaptation is a classification problem that uses labeled source domain to help unlabeled target domain achieve better classification results. The main approach to unsupervised domain adaptation is to make classifiers trained on the source domain and use on the target domain by reducing the domain discrepancy between the source and target domains. However, most of the current domain adaptation methods treat the marginal and conditional probability distribution equally, In practical applications, since the discrepancy between the source and target domains are various, it is necessary to dynamically adjust the weight parameters of the source and target domain adaptation. In this paper, we propose a method to dynamically adapt the probability distributions of source and target domains, and learn the class discriminative features in the adaptation process to constrain the distance between samples of the same category to be as small as possible and the distance between samples of different categories to be as large as possible in the shared common subspace, so that the original samples have better differentiability after the projection. A large number of experiments have conducted on several visual cross-domain classification tasks show that our algorithm behaves much better than the previous ones.