[http://citeseer.nj.nec.com/ku95cellular.html]

by KimWCKu, MWMak, WCSiu (1995)

abstract: ["RecurrentNeuralNetwork"]s (RNNs), with the capability of dealing with spatio-temporal relationship, are more complex than feed-forward NeuralNetworks. Training of RNNs by GradientDescent methods becomes more difficult. Therefore, another training method, which uses ["CellularGeneticAlgorithm"]s, is proposed. In this paper, the performance of training by a GradientDescent method is compared with that by a CellularGeneticAlgorithm. Experimental results indicate that the cellular genetic...

refers to GradientDescent against an ErrorFunction training algorithms in (7,8,10)

refers to more information on ["GeneticAlgorithm"]s in (4)

refers to CellularGeneticAlgorithm""s in (9) (training the RNN with predefined topologies)

compares ["Robust"]ness and generalization capabilities of ["CellularGeneticAlgorithm"]s versus standard RealtimeRecurrentLearning (algorithm) (RTRL referenced further in (10) ).