Journal of Guangdong University of Technology ›› 2007, Vol. 24 ›› Issue (4): 38-41.

• Comprehensive Studies • Previous Articles     Next Articles

A New ICA Algorithm Based on Normalized Kurtosis

  

  1. (Faculty of Applied Mathematics,Guangdong University of Technology,Guangzhou 510006,China)
  • Online:2007-11-08 Published:2007-11-08

HTML

PDF (PC)

2318

Abstract: The traditional approach to gradient to search for the cost function’s extreme in the independent component analysis was based on kurtosis.The path we are looking for is crucial to the convergence of the algorithm.In the Kuicnet algorithm,after replacing the gradient of the cost function with its scaling we obtained a new ICA algorithm.Adjusting the path of the gradient timely in the iterative process,it accelerates the convergence.The simulation illustrates the effectiveness.

Key words: independent component analysis； kurtosis； fast-ICA algorithm；

[1] DOUGLAS S C,KUNG S Y.Kuicnet algorithms for blinddecon-volution. IEEE Workshop Neural Networks Sig-nal Process Cambridge . 1998

[2] HYVARINEN A,OJA E.Independent Component Analy-sis:Algorithms and Application. Neural Networks Cen-tre . 2000

[3] GEORGIE P,CICHOCKI A,AMARI S.On some exten-sions of the natural gradient algorithm. Third Interna-tional Conference on Independent Component Analysis andBlind Signal Separation(ICA 2001) . 2001

[4] HYVARINEN A,KARHUNEN J,OJA E.Independentcomponent analysis. . 2001

[5] JUTTEN C.Blind separation of sources. Signal Processing . 1991

[6] COMON P.Independent component analysis,a new con-cept. Signal Processing . 1994

[7] Lee T W.Independent component analysis using an extended infomax algorithm for mixed subgaussion and supergaussion sources. Neural Computation . 1999

[8] Hyv rinen A.Fast and robust fixed-point algorithms for independent component analysis. IEEE Transactions on Neural Networks . 1999
No related articles found!
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
Copyright © 2017 Journal of Guangdong University of Technology
Add：729 East Dongfeng RD., Guangzhou PRC. Postcode: 510090
Tel：020-37626653,020-37626139,020-37626396
Email：xbzrb@gdut.edu.cn
Supported:Beijing Magtech