PROJECT TITLE :

Adaptive Local Embedding Learning for Semi-supervised Dimensionality Reduction

ABSTRACT:

In recent years, a lot of attention has been focused on semi-supervised learning, which is recognized as one of the most appealing problems in the machine learning research field. In this paper, we propose a novel locality preserved dimensionality reduction framework. We call it Semi-supervised Adaptive Local Embedding learning (SALE), and it learns a local discriminative embedding by constructing a k1 Nearest Neighbors ( k1 NN ) graph on labeled data. This is done in order to explore the intrinsic structure, also known as sub-manifolds, that are present in non-Gaussian labeled data. The next step is to map all of the samples into the learned embedding and build a new k2 NN graph on all of the embedded data in order to investigate the overall structure of all of the samples. Therefore, the unlabeled data and their corresponding labeled neighbors can be clustered into the same sub-manifold to improve the discriminative power of embedded data. This is done in order to maximize the information gained from the embedded data. In addition, based on the proposed SALE framework, we present two semi-supervised dimensionality reduction methods with orthogonal and whitening constraints. In order to solve the NP-hard problem that is present in our models, an effective alternatively iterative optimization algorithm has been developed. Our methods have been shown to be superior for the exploration and classification of local structures, as demonstrated by extensive experiments carried out on a variety of synthetic and actual world data sets.