Low-Rank Transfer Learning and Its Application
For knowledge-based machine learning algorithms, label or tag is critical in training the discriminative model. However, labeling data is not an easy task because these data are either too costly to obtain or too expensive to hand-label. For that reason, researchers use labeled, yet relevant, data from different databases to facilitate learning process. This is exactly transfer learning that studies how to transfer the knowledge gained from an existing and well-established data (source) to a new problem (target). To this end, we propose a method to align the structure of the source and target data in the learned subspace by minimizing the reconstruction error, called low-rank transfer subspace learning (LTSL). The basic assumption is if each datum in a specific neighborhood in the target domain can be reconstructed by the same neighborhood in the source domain, then the source and target data might have similar distributions. The benefits of this method are two-fold: (1) generality to subspace learning methods, (2) robustness by low-rank constraint. Extensive experiments on face recognition, and objection recognition demonstrate the effectiveness of our method.