Computational Intelligence

Implementation Issues 203

Some implementations choose values in the range 0 to 1. Other implementations

generate initial weight vectors lying at random locations on the unit hypersphere.

The weight vector normalization procedure is done for all weights connected to a

given output processing element, from all input processing elements. The most logi-

cal way to do this would seem to be to set the square root of the sum of the squares of

the weights from all of the inputs to each output to the same value, presumably 1. The

reason we say “would seem” and “presumably” is that various examples of Kohonen

implementations have normalized weights in different ...