Kernel-independent component analysis explained

In statistics, kernel-independent component analysis (kernel ICA) is an efficient algorithm for independent component analysis which estimates source components by optimizing a generalized variance contrast function, which is based on representations in a reproducing kernel Hilbert space.[1] [2] Those contrast functions use the notion of mutual information as a measure of statistical independence.

Main idea

Kernel ICA is based on the idea that correlations between two random variables can be represented in a reproducing kernel Hilbert space (RKHS), denoted by

l{F}

, associated with a feature map

Lx:l{F}\mapstoR

defined for a fixed

x\inR

. The

l{F}

-correlation between two random variables

X

and

Y

is defined as

\rhol{F

}(X,Y) = \max_ \operatorname(\langle L_X,f \rangle, \langle L_Y,g \rangle)

where the functions

f,g:R\toR

range over

l{F}

and

\operatorname{corr}(\langleLX,f\rangle,\langleLY,g\rangle):=

\operatorname{cov
(f(X),

g(Y))}{\operatorname{var}(f(X))1/2\operatorname{var}(g(Y))1/2}

for fixed

f,g\inl{F}

. Note that the reproducing property implies that

f(x)=\langleLx,f\rangle

for fixed

x\inR

and

f\inl{F}

.[3] It follows then that the

l{F}

-correlation between two independent random variables is zero.

This notion of

l{F}

-correlations is used for defining contrast functions that are optimized in the Kernel ICA algorithm. Specifically, if

X:=(xij)\inRn

is a prewhitened data matrix, that is, the sample mean of each column is zero and the sample covariance of the rows is the

m x m

dimensional identity matrix, Kernel ICA estimates a

m x m

dimensional orthogonal matrix

A

so as to minimize finite-sample

l{F}

-correlations between the columns of

S:=XA\prime

.

Notes and References

  1. Bach . Francis R. . Jordan . Michael I. . 10.1162/153244303768966085 . Kernel independent component analysis . The Journal of Machine Learning Research . 3 . 1–48 . 2003 .
  2. Book: Bach . Francis R. . Jordan . Michael I. . 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03) . Kernel independent component analysis . 10.1109/icassp.2003.1202783 . 4 . IV-876-9 . 2003 . 978-0-7803-7663-2 . 7691428 .
  3. Book: Saitoh, Saburou . Theory of Reproducing Kernels and Its Applications . Longman . 1988. 978-0582035645.