Concise Coupled Neural Network Algorithm for Principal Component Analysis

Authors:
Lijun Liu;Jun Tie;Tianshuang Qiu
Affiliations:
School of Electronic and Information Engineering, Dalian University of Technology, Dalian, China 116024 and Department of Mathematics, Dalian Nationalities University, Dalian, China 116605;Department of Mathematics, Dalian Nationalities University, Dalian, China 116605;School of Electronic and Information Engineering, Dalian University of Technology, Dalian, China 116024
Venue:
ISNN '08 Proceedings of the 5th international symposium on Neural Networks: Advances in Neural Networks
Year:
2008

Citing 8
Cited 0

Original Contribution: Principal components, minor components, and linear neural networks

Neural Networks
Principal component neural networks: theory and applications

Principal component neural networks: theory and applications
Projection approximation subspace tracking

IEEE Transactions on Signal Processing
Robust recursive least squares learning algorithm for principal component analysis

IEEE Transactions on Neural Networks
Coupled principal component analysis

IEEE Transactions on Neural Networks
Robust principal component analysis by self-organizing rules based on statistical physics approach

IEEE Transactions on Neural Networks
Principal component extraction using recursive least squares learning

IEEE Transactions on Neural Networks
An adaptive learning algorithm for principal component analysis

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

A concise ordinary differential equations (ODE) for eigen-decomposition problem of a symmetric positive matrix is proposed in this paper. Stability properties of the proposed ODE is obtained by the theory of first order approximation. Novel coupled neural network (CNN) algorithm for principal component analysis (PCA) is obtained based on this concise ODE model. Compared with most non-coupled neural PCA algorithms, the proposed online CNN algorithm works in a recursive manner and simultaneously estimates eigenvalue and eigenvector adaptively. Due to the fact the proposed CNN effectively makes use of online eigenvalue estimate during learning process, it reaches a fast convergence speed, which is further verified by the numerical experiment result. Adaptive algorithm for sequential extraction of subsequent principal components is also obtained by means of deflation techniques.