An Evolutionary Algorithm for the Generation of Unsupervised Self Organising Neural Nets.
Tim Hendtlass · Industrial and Engineering Applications of Artificial Intelligence and Expert Systems · 2022
An evolutionary algorithm is described that allows the speedy generation of unsupervised self organising maps. The maps produced are memory efficient in that they use almost the minimum number of nodes required to hold a specified level of detail about the training set. This level of detail can be explicitly controlled. Compared to conventional SOM techniques, the one described is faster and produces nets with far smaller numbers of nodes. The algorithm, although described here in terms of generating three dimensional nets, is applicable to the generation of nets of any desired dimensionality.