ica - Independent Component Analysis Mixture Data Data that...

Info iconThis preview shows pages 1–8. Sign up to view the full content.

View Full Document Right Arrow Icon
Independent Component Analysis Independent Component Analysis
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
2 Mixture Data x Data that are mingled from multiple sources b May not know how many sources b May not know the mixing mechanism x Good Representation b Uncorrelated, information-bearing components h PCA and Fisher’s linear discriminant b De-mixing or separation h ICA (Independent component analysis) x How do they differ?
Background image of page 2
3 PCA vs. ICA x Independent events vs. Uncorrelated events Knowing X1 doesn’t tell anything about X2 Knowing X1 does tell something about X2 x1 x2 x2
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
4 Uncorrelated vs. Independence x Uncorrelated b Global property b Not valid under nonlinear transform b PCA requires uncorrelation x Independence b Local property b Valid for nonlinear transform b ICA assumes independence 0 )) )( (( : )) ( ( )) ( ( )) ( , ), ( ), ( ( : 2 2 1 1 1 1 2 2 1 1 = - - = Ex x Ex x E ed uncorrelat g x g E x g E x g x g x g E ce independen n n n n L L
Background image of page 4
5 Uncorrelated vs. Independence x Independence is stronger, requiring every possible function of x1 to be uncorrelated with x2 x E((y1-E(y1))(y2-E(y2))=0 -> uncorrelated x y2= y1 2 -> not independent
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
6 Uncorrelated vs. Independence x Discrete variables X1 and X2 x (0,1), (0,-1),(1,0),(-1,0) all with ¼ probability x X1 and X2 are uncorrelated x E(x1 2 x2 2 )=0!=1/4=E(x1 2 )E(x2 2 )
Background image of page 6
7 ICA Limitation x Any symmetrical distribution of x1 and x2 around origin (centered at Ex1 and Ex2) is uncorrelated x Corollary: ICA does not apply to Gaussian
Background image of page 7

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 8
This is the end of the preview. Sign up to access the rest of the document.

Page1 / 22

ica - Independent Component Analysis Mixture Data Data that...

This preview shows document pages 1 - 8. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online