|
|
|
| Dimensionality Reduction and Reconstruction of Data Based on Autoencoder Network |
| Hu Zhao-hua①②; Song Yao-liang① |
| ①School of Electronic Engineering and Optoelectronic Technology, Nanjing University of Science and Technology, Nanjing 210094, China;②College of Electronic & Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, China |
|
|
|
|
Abstract The curse of dimensionality is a central difficulty in many fields such as machine learning, pattern recognition and data mining etc. The dimensionality reduction method of characteristic data is one of the current research hotspots in data-driven calculation methods, which high-dimensional data is mapped into a low-dimensional space. In this paper, a special nonlinear dimensionality reduction method called “Autoencoder” is introduced, which uses Continuous Restricted Boltzmann Machine (CRBM) and converts high-dimensional data to low-dimensional codes by training a neural network with multiple hidden layers. In particular, the “autoencoder” provides such a bi-directional mapping between the high-dimensional data space and the low-dimensional manifold space and is therefore able to overcome the inherited deficiency of most nonlinear dimensionality reduction methods that do not have an inverse mapping. The experiments on synthetic datasets and true image data show that the “autoencoder” network not only can find the embedded manifold of high-dimensional datasets but also reconstruct exactly the original high-dimension datasets from low-dimensional structure.
|
|
Received: 22 April 2008
|
|
|
|
|
|
|
|
|
2009, Vol. 31 
