Architecture of neural networks using genetic algorithms

Neural computing is one of the most rapidly expanding areas of currentresearch, attracting researchers from a wide variety of disciplines. The main cause of this popularity is that the approach of neural computing tries to capturethe guiding principles that underly the brain solution to complex problems andapplies them to computer systems. The above reason makes artificial neuralnetworks a very appealing choice in solving a diverse number of complexproblems in which other techniques fail to provide satisfactory solutions.One of the most significant development in the area of neural networks is thebackpropagation algorithm. But the use of backpropagation algorithm to trainfeedforward neural networks assumes that the structure of the networks and thevalues of learning rule parameters are predetermined before starting the trainingof the network. In most cases there is no simple method to predetermine theoptimal structures and the best learning parameters. The choice of these values isusually done by trial-and-error method of different combinations of .theparameter values until a desired performance is obtained. However, this is atedious and computationally expensive process.This thesis introduces a new algorithm in order to select systematically theproper network architectures. The new algorithm was developed by integrating agenetic algorithm with a momentwn backpropagation learning algorithm. Thegenetic algorithm is used to optimize network architectures and thebackpropagation algorithm is used to train networks and generate connectionweights. The developed algorithm is used to alleviate the burden ofpredetermining the network architecture and learning rule parameters byautomatically searching the space of possible network architectures. The bestnetwork architecture is evolved with the associated learning rule parameters