To be able to build modular neural networks where the
output is a weighted value of the outcomes of several
trainded networks, we need a mechanism to build a
confusion matrix for a trained neural network. To do
this we need to extend the TeacherSynapse object, I think.