A discriminative feature mapping approach to heterogeneous domain adaptation

Heterogeneous domain adaptation algorithms address a key issue in applied machine learning: How can we improve the learning task in one domain by leveraging the data from another domain? We extend this problem to the more severe case where the two input domains do not share any common features or instances. In this paper, we present a simple and intuitive technique called Cross-Domain Mappings (CDM) to address the problem. First, we separate the source data instances into distinct classes by using a separation method. We then apply a regression technique to map each labeled target data instance to be as close as possible to the center of the source data group with the same class label. Finally, we again use a separation method to separate all the data instances into distinct classes. Experimental results on some benchmark data sets clearly demonstrate that our approach is effective for learning discriminative features of supervised classification with few training instances from the target domain in practice.