Domain adaptation problems arise in a variety of applications, where atraining dataset from the \textit{source} domain and a test dataset from the\textit{target} domain typically follow different distributions. The primarydifficulty in designing effective learning models to solve such problems liesin how to bridge the gap between the source and target distributions. In thispaper, we provide comprehensive analysis of feature learning algorithms used inconjunction with linear classifiers for domain adaptation. Our analysis showsthat in order to achieve good adaptation performance, the second moments of thesource domain distribution and target domain distribution should be similar.Based on our new analysis, a novel extremely easy feature learning algorithmfor domain adaptation is proposed. Furthermore, our algorithm is extended byleveraging multiple layers, leading to a deep linear model. We evaluate theeffectiveness of the proposed algorithms in terms of domain adaptation tasks onthe Amazon review dataset and the spam dataset from the ECML/PKDD 2006discovery challenge.
展开▼
