Assume the set training patterns be the , the set training patterns from .

Where

and

Classify (a point in feature space) as coming from if the average distance from x to training patterns from is less than the average distance from x to training patterns from .

Where

Let:

The right and left terms in the upper equation represent the squared Euclidean distance between d-vectors and .

We can derive the relation between g(x) and W as follows:

As shown in Fig.6, we can obtain:

As shown in Fig.6, T1 and T2 are called templates (Class centroid). If

then .

index