A Novel Dimensionality Reduction Framework Based on Neural Network

• 1、School of Economics and Management, Beijing University of Posts and Telecommunications, Beijing 100876

Abstract：Dimensionality reduction is a direct and effective method for analyzing and visualizing high-dimensional data. Although there are a variety of data dimensionality reduction methods, these methods have not been reasonably and effectively unified. This paper presents a generic and scalable dimensionality reduction framework(DRNN) based on Neural Network, which designs a generic cost function form according to the basic principle of the dimensionality reduction, so that the neural network can be used to better realize the original data dimensionality reduction. Experimental results show that the DRNN framework can achieve the same or better dimensionality reduction performance than the original method, and can effectively solve the out-of-sample problems existing in traditional dimensionality reduction method.

Keywords： Neural network Data dimension reduction Cost function Dimensionality reduction generalized framework