Domain Invariant and Class Discriminative Heterogeneous Domain Adaptation

Heterogeneous domain adaptation (HDA) aims to improve the task of target domain by source domain, despite the data distribution of source and target domains are different. In our algorithm, we match the marginal and condition distribution, and introduce the repulsive force term to expand the difference between classes. Finally we find a latent subspace by learning different transformations for different domains. At the same time, we add structural consistency constraints and l 2,1 -norm to these transformations, making them more conducive to feature selection. Extensive experiments verify that our algorithm can significantly outperform several state-of-the-art methods on several datasets of image classification.