A generic algorithm for training networks with artificial dendritic trees
John G. Elias, Baoming Chang · 2003
A specialized genetic algorithm for training artificial neural networks which are constructed from artificial dendritic trees and their collection of artificial synapses is described. It is shown that artificial neural networks with dendritic tree structures can be trained by changing their connections to sensory devices, e.g., CCD (charge coupled device) arrays, and connections to other artificial neurons. The number of different connection patterns is a combinational problem which grows factorially as the number of artificial synapses in the network and the number of sensor elements increase. It is shown that a specialized genetic algorithm produces promising results for a simple application using these types of networks. It is found that the crossover operator works well operating on connections rather than bit strings and that an embedded optimizer in place of the mutation operator greatly improves training performance.>