New paper: Relative Newton Method for Quasi-ML Blind Source Separation

[Michael Zibulevsky]

Announcing a paper ...

Title:   Relative Newton Method for Quasi-ML Blind Source Separation

Author:  Michael Zibulevsky 

ABSTRACT:

Presented relative Newton method for quasi-maximum likelihood blind source
separation significantly outperforms  natural gradient descent in batch mode.
The structure  of the corresponding Hessian matrix allows its fast inversion
without assembling. Experiments with sparsely representable signals and images
demonstrate super-efficient separation.


URL of gzipped ps file:

http://ie.technion.ac.il/~mcib/newt_ica_jmlr1.ps.gz

Contact:  mzib at ee.technion.ac.il


===========================================================================
Michael Zibulevsky, Ph.D.                  Email: mzib at ee.technion.ac.il 
Faculty of Electrical Engineering          Phone: 972-4-829-4724 
Technion - Israel Institute of Technology         972-4-832-3885 
Haifa 32000, Israel                        Cell:  972-55-968297
http://ie.technion.ac.il/~mcib/            Fax:   972-4-829-4799
===========================================================================