After successful learning it can be particularly interesting whether the network has only memorized i.e. whether it is able to use our training trials to quite exactly produce the suitable output but to provide wrong answers for everybody other difficulties of the same class.
Suppose that we would like the network to train a mapping and therefore use the training samples from fig. 4.1: Then there can be a chance that, lastly, the network mwill precisely mark the colored locations around the training samples with the output and other-wise will output 0. Thus, it includes sufficient storage capability to concentrate on the six trainingsamples with the output 1. This impliesan bulky network with too much freestorage capacity.

Then again a network might have insufficient capacity this bad presentation of input data will not correspond to the good generalization performance we desire. Thus, we have to find the balance.