Computational Intelligence - Learning with Neural Methods on Structured Data

Research of Marc Strickert

The animation shows a standard benchmark for SOM: Adaptation of a square neuron lattice to a given set of points. Starting with a fast ordering phase of the random location initialization, the annealing of the global learning parameter leads to stagnation, covering the given set of 218 points with only 100 neurons reasonably well after 850 learning cycles.