The Combination of Genetic Programming and Genetic Algorithm for Neural Networks Design and Training

Abstract

We present in this paper using of two evolutionary computations tools for design and training feed-forward neural networks . We use genetic programming algorithm to discover suitable design for neural network that modeled the selected problem . This discover design it have features that make neural network less cost in structure (the smaller network topology) that give desired output for problem . Genetic programming it is a search algorithm that dell with population of tree structures , each of these tree structure it use as suggested design for neural network . The optimization is done throw minimization number of hiding layers and number of neurons in each layer with less connectivity with other neurons . Each new generated neural net send to genetic algorithm for training . Genetic algorithm work as learning algorithm that specify the training set of weights that linked with neural network connection . This work represent global approach that give promising result for some problems .