Feature Transformation

The method of PDF projection (without the MaxEnt constraint) has existed since 2000 [4,5]. We consider an arbitrary feature transformation that maps from ${\cal R}^N$ to ${\cal R}^D$, where $D \leq N$ :

$${\bf z}=T({\bf x}), \;\;\; {\bf x}\in{\cal R}^N, \;\;\; {\bf z}\in{\cal R}^D.$$ (2.1)