Convergence of algorithms used for principal component analysis
- 40 Downloads
The convergence of algorithms used for principal component analysis is analyzed. The algorithms are proved to converge to eigenvectors and eigenvalues of a matrixA which is the expectation of observed random samples. The conditions required here are considerably weaker than those used in previous work.
Keywordsprincipal component analysis stochastic approximation algorithms convergence
Unable to display preview. Download preview PDF.
- 1.Oja, E.,Subspace Methods of Pattern Recognition, 1st ed., Letchworth, Hertfordshire: Research Studies Press Ltd., 1983.Google Scholar
- 2.Deprettere, E. F. (eds.),SVD and Signal Processing, North-Holland: Elsevier, 1988.Google Scholar
- 3.Vaccaro, R. (eds.),SVD and Signal Processing, II:Algorithms, Analysis and Applications, Amsterdam: Elsevier, 1991.Google Scholar
- 7.Kung, S. Y., A new identification and model reduction algorithm via singular value decomposition, inProc. 12th Asilomar Conf. Circuits, Syst. Comput., Pacific Grove, Calif., 1978 (eds. Chi-Chia Hsieh), New York: IEEE Computer Society, 1979, 705–714.Google Scholar
- 8.Kushner, H. J., Clark, D. S.,Stochastic Approximation Methods for Constrained and Unconstrained Systems, 1st ed., New York: Springer-Verlag, 1978.Google Scholar
- 9.Chen, H. F., Zhu, Y. M.,Stochastic Approximation (in Chinese), 1st ed., Shanghai: Shanghai Science and Technology Press, 1996.Google Scholar
- 10.Chen, H. F., Stochastic approximation and its new applications,Proc. International Workshop on New Directions of Control and Manufacturing, Hong Kong, 1994, 2–12.Google Scholar