Classifier Topologies

Up to now, we have discussed PDF projection and MaxEnt PDF projection only from the point of view of the feature transformation, $T({\bf x})$, computing the projected PDF $G({\bf x};H_0,T,g)$ or the MaxEnt PDF $G^*({\bf x}; T,g)$ and drawing random samples from the same. These are essentially the building-blocks needed to create classifiers. In this chapter, we discuss various means of creating classifiers from these building building blocks. The various methods are distinguished by (a) how the data is segmented, (b) which window functions are used, and (c) how the various likelihood functions are combined. We may discuss segentation and window functions (a,b) independently of likelihood functions combination (c).