Description: FAST KERNEL ICA |
|--------------------|
Version 1.0- February 2007
Copyright 2007 Stefanie Jegelka, Hao Shen, Arthur Gretton
This package contains a Matlab implementation of the Fast Kernel ICA
algorithm as described in [1].
Kernel ICA is based on minimizing a kernel measure of statistical
independence, namely the Hilbert-Schmidt norm of the covariance
operator in feature space (see [3]: this is called HSIC). Given an (n
x m) matrix W of n samples from m mixed sources, the goal is to find a
demixing matrix X such that the dependence between the estimated
unmixed sources X *W is minimal. FastKICA uses an approximate Newton
method to perfom this optimization. For more information on the
algorithm, read [1], and for more information on HSIC, refer to [3].
The functions chol_gauss and amariD are taken from and based on,
respectively, code from Francis Bach (available at
http://cmm.ensmp.fr/~bach/kernel-ica/index.htm). The derivative is
com
File list (Check if you may need any files):
fastKICA\amariD.m
........\demo.m
........\fastkica.m
........\Readme (2).txt
........\README.txt
........\source2.wav
........\source3.wav
........\source4.wav
........\utils\chol_gauss.c
........\.....\compDerivChol.m
........\.....\dChol.m
........\.....\dChol2.c
........\.....\dChol2Lin.c
........\.....\dCholLin.m
........\.....\dKmn.c
........\.....\dKmnLin.c
........\.....\getKern.c
........\.....\hessChol.m
........\.....\hsicChol.m
........\utils
fastKICA
Download