A neural network is formed by a network of computing units (the neurons)
linked to each other. Each of these connections has an associated weight. Con-
structing a neural network consists of using an algorithm to find the weights of
the connections between the neurons. A neural network has its neurons orga-
nized in layers. The first layer contains the input neurons of the network. The
cases of the problem we are addressing are presented to the network through
these input neurons. The final layer contains the predictions of the neural net-
work for the case presented at its input neurons. In between we have one or
more “hidden” layers of neurons. The weight updating algorithms, like for in-
stance the back-propagation method, try to obtain the connection weights that
optimize a certain error criterion, that is the weights which ensure that the
network output is in accordance to the cases presented to the model.