In search of a good neuro-genetic computational paradigm

This paper reports the effect of some advanced genetic operators like two-parents multipoint restricted crossover (Double-MRX), three-parents multipoint restricted crossover (Triple-MRX), elitist selection and scheduled mutation on the adaptability of feedforward neural networks trained over complex and computationally expensive electronic nose data. The authors show that the performance of Triple-MRX is better that of Double-MRX. Upon applying elitist selection with Double-MRX and scheduled mutation with Triple-MRX, the performance of the genetic training of the neural network improves up to some extent, but Triple-MRX is still better than Double-MRX as far as quality of solution and speed of convergence are concerned. It is also shown that the performance levels of these hybrid techniques far exceeds those of the commonly used backpropagation model. The search for a good neuro-genetic hybrid computational paradigm based on advanced genetic operators is a frontier research area in the evolution of a sixth generation computing system.