Class-Specific Reconstruction Transfer Learning via Sparse Low-Rank Constraint

Shanshan Wang, Lei Zhang, Wangmeng Zuo; Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 949-957

Subspace learning and reconstruction has been widely explored in transfer learning. However, existing subspace reconstruction neglect class prior such that the learned transfer function is biased. We propose a novel reconstruction-based method called Class-specific Reconstruction Transfer Learning (CRTL), which optimizes a well-designed transfer loss function without class bias. Using a class-specific reconstruction matrix to align source domain with target domain which provides help for classification with class prior modeling. Furthermore, to keep the intrinsic relationship between data and labels after feature augmentation, a projected HSIC, that measures the dependency between two sets, is first proposed by mapping the data from original space to RKHS. In addition, combining low-rank and sparse constraints on reconstruction matrix, the global and local data structures can be effectively preserved. Extensive experiments demonstrate our method outperforms conventional methods.