Learning the Supervised NLDR Mapping for Classification

Abstract

In researches and experiments, we often work with large volumes of high-dimensional data and regularly confront the problem of dimensionality reduction. Some Non-Linear Dimensionality Reduction (NLDR) methods have been developed for unsupervised datasets. As for supervised datasets, there is a newly developed method called SIsomap, which is capable of discovering structures that underlie complex natural observations. However, SIsomap is limited from the fact that it doesn’t provide an explicit mapping from original space to embedded space, and thus can’t be applied to classification. To solve this problem, we apply neural network to learning that mapping and then to classify. We test our method on a real world dataset. To prove the effectiveness of our method, we also compare it with a related classification method, Extended Isomap. Experiments show that our proposed method has satisfactory performance.