DiReT: An effective discriminative dimensionality reduction approach for multi-source transfer learning

Transfer learning is a well-known solution to the problem of domain shift in which source domain (training set) and target domain (test set) are drawn from di erent distributions. in the absence of domain shift, discriminative dimensionality reduction approaches could classify target data with acceptable accuracy. however, distribution di erence across source and target domains degrades the performance of dimensionality reduction methods. in this paper, we propose a discriminative dimensionality reduction approach for multi-source transfer learning, diret, in which discrimination is exploited on transferred data. diret nds an embedded space, such that the distribution di erence of the source and target domains is minimized. moreover, diret employs multiple source domains and semi-supervised target domain to transfer knowledge from multiple resources, and it also bridges across source and target domains to nd common knowledge in an embedded space. empirical evidence of real and arti cial datasets indicates that diret manages to improve substantially over dimensionality reduction approaches.