Semi-Supervised Locality Discriminant Projection

Abstract

In this paper, we consider the problem of semi-supervised dimensionality reduction. We focus on the local geometric structure of data and propose a novel method, called Semi-supervised Locality Discriminant Projections (SSLDP). It uses both labeled and unlabeled samples. Specifically, the labeled samples are used to explore the discriminating ainformation including both similarity and dissimilarity information, while the unlabeled samples are used to estimate the intrinsic geometric structure of data. Thus, SSLDP learns a discriminant projection which can best preserve both the discriminating structure and the local geometric structure of data. We evaluate SSLDP in the similarity measure which plays a key role in most of the information processing tasks. The experimental results show the effectiveness of our algorithm.