Knowledge-Based Systems

208 Chapter 7 ConneCtionist Models

end of training, only the winning unit is adjusted. Similarly, the learning rate

will decrease as training progresses, and in some implementations, the learn-

ing rate decays with the distance from the winning output unit.

A basic Kohonen algorithm can be given as follows:

begin

randomize weights for all neurons

for (i = 1 to iteration_number) do

begin

take one random input pattern

find the winning neuron

find neighbors of the winner

modify synaptic weights of these neurons

reduce the η (learning rate) and λ (neighborhood radius)

end

end.

Figure 7.10 represents the basic structure and steps of Kohonen ...