The neurons with high activity (i.e. neurons which "prototype"
signal is similar to input signal) are the most relevant during the learning
phase. An competitive and iterative algorithm is used in order to extract
neurons that are locally relevant. We define local interactions (typically Mexican
hat function) that a neuron creates in its neighborhood.
In the
following picture, a high activity neuron is represented by a white square, a
low activity neuron by a black one.