Sparsely Preserving Based Semi-supervised Dimensionality Reduction

Abstract

Sparse subspace learning has drawn more and more attentions recently. However, most of the sparse subspace learning methods are unsupervised or supervised, which cannot utilize the information offered by unlabeled data. In this paper, a new sparse subspace learning algorithm called Sparsely Preserving Based Semi-Supervised Dimensionality Reduction (SPSSDR) is proposed by adding the sparsely information into liner discriminant analysis. SPSSDR not only preserves the sparse relationship among samples, but also preserves the category structure of data; furthermore, it also utilizes some information offered by unlabeled data. The conducted experiments on challenging benchmark datasets validate the SPSSDR.