Terminal layer of a CeNiN network. It does not compute activations; it exposes the class labels together with the probabilities produced by the preceding softmax layer.