Optimal Hardware Implementation of a Feedforward Neural Network Topology using a Genetic Algorithm for Prunning

The genetic algorithms are an important part of modern global searching methods class with specific applications in the complex optimization problems. This paper proposes an interesting approach of feedforward neural network topology optimization based on a new fitness function definition. The experimental results getting in this case are compared with ones from a classic pruning method, and for their validation a proper hardware implementation of the used networks is indicated.