Whitening

From Ufldl

Jump to: navigation, search
(ZCE Whitening)
Line 90: Line 90:
of the data, and do not try to reduce its dimension.  
of the data, and do not try to reduce its dimension.  
-
\smallskip
+
== Regularizaton ==
-
\noindent
+
When implementing PCA whitening or ZCA whitening in practice, sometimes some
-
'''Regularization.''' When implementing PCA whitening or ZCA whitening in practice, sometimes some
+
of the eigenvalues <math>\textstyle \lambda_i</math> will be numerically close to 0, and thus the scaling
of the eigenvalues <math>\textstyle \lambda_i</math> will be numerically close to 0, and thus the scaling
step in Equation~(\ref{eqn-sqrtlambda}) above would involve dividing by a value close to zero, and may cause
step in Equation~(\ref{eqn-sqrtlambda}) above would involve dividing by a value close to zero, and may cause
Line 104: Line 103:
When <math>\textstyle x</math> takes values around <math>\textstyle [-1,1]</math>, a value of <math>\textstyle \epsilon \approx 10^{-5}</math>
When <math>\textstyle x</math> takes values around <math>\textstyle [-1,1]</math>, a value of <math>\textstyle \epsilon \approx 10^{-5}</math>
might be typical.  With this form of regularization, the features won't all
might be typical.  With this form of regularization, the features won't all
-
have exactly variance 1, but they will still be uncorrelated.  We then
 
-
similarly compute the ZCA whitened data as <math>\textstyle x_{\rm ZCAwhite} = U x_{\rm PCAwhite}</math>.
 
-
 
-
[[TODO: Double check default/reasonable value of <math>\textstyle \epsilon</math>]]
 
For the case of images, adding <math>\textstyle \epsilon</math> here also has the effect of slightly smoothing (or low-pass
For the case of images, adding <math>\textstyle \epsilon</math> here also has the effect of slightly smoothing (or low-pass

Revision as of 11:09, 4 April 2011

Personal tools