Footnote
For those who know linear algebra and are interested, here is a sketch of
the proof.
Let feature vector x have mean mx
and covariance matrix Cx, and let feature vector y
have mean my and covariance matrix Cy.
If y = Ax, we know that my
= Amx and Cy = A Cx A'.
From this last expression, it follows that Cy-1 =
A-1 ' Cx-1 A-1. Now consider
ry2 = ( y - my
)' Cy-1 ( y - my
)
= ( Ax - Amx )'
( A-1 ' Cx-1 A-1 ) ( Ax
- A mx )
= ( x - mx )' A' A-1
' Cx-1 A-1 A ( x - mx
)
= ( x - mx )' (A-1
A)' Cx-1 (A-1 A) ( x -
mx )
= ( x - mx )' Cx-1
( x - mx )
= rx2 .
Back