A problem that occasionally arises in the application of the SOM algorithm is the failure of topological ordering by creating a "folded" map this problem arises when the neighborhood size is permitted to decay too rapidly. The creation of a folded map may be viewed as some form of a "local minimum" of the topological-ordering process
To investigate this phenomenon, consider a two-dimensional lattice of a 10 X 20 network of neurons trained on a two-dimensional input uniformly distributed inside the square
Compute the map produced by the SOM algorithm, permitting the neighborhood function around the winning neuron to decay much faster than that normally used. You may have to repeat the experiment several times in order to see a failure of the ordering process.