Evolutionary approach to design of Artificial Neural Networks

Nowadays several techniques of optimization of neural networks have being researched, one of them use evolutionary computation. The objective of this research is to introduce an. Evolutionary system biologically plausible, as far as possible that can automatically generate Artificial Neural Networks (ANN) with good generalization capacity, smaller error and larger tolerance to noises. To this aid, three biological metaphors were used: Genetic algorithms (GA), Lindenmayer Systems (L-System) and ANN. First it was introduced the biological metaphors used in this research at the end the results of simulation for the parity problem are presented. The method is better than the other ones because it increases the level of implicit parallelism of genetic algorithm and for the aspects of biological plausibility. The system generates the minimum satisfactory architecture that solves a specific task, reducing the project costs and increasing the performance of the neural networks obtained.