Dimension Reduction Algorithm of Big Data Based on Deep Neural Network

Abstract

Dimensionality reduction refers to mapping data from high-dimensional space to a low-dimensional space, which is the main method of high-dimensional data mining. The restricted Boltzmann machine, the basic structure module of deep learning, is introduced and analyzed. The high-dimensional data reduction model based on restricted Boltzmann machine is established. At the same time, the validity has been verified theoretically and experimentally. It makes the depth model as compact and simple as possible without losing the precision, and the calculation speed is improved. Experiment results show that the model identification is of high accuracy. In particular, the depth model built by RBM based on the adaptive adjustment of hidden layer node units not only has high recognition accuracy, but also greatly improves the computing speed and saves the storage space.