In a simple competitive network the weight vectors of the nodes in the output layer of
the network after training should ideally represent the prototypes of the clusters found
in the data set. The initial values for these prototypes - the weight vectors - are
normally set to random values in an attempt to avoid bias. Each input from the data set
is then presented to the network in turn and output layer nodes compete for the right to
classify the input and update their weight vectors. The winner of the competition is the
node whose weight vector is most similar to the input vector presented. If the winning
node's weight vector is wi* then: