Linear Network Representation Applied to evolving of the Architecture and Weights of Neural Network

Abstract

Evolutionary computation is a class of global search techniques based on
the learning process of a population of potential solutions to given problem that
has been successfully applied to a variety of problems. In this paper a new
approach to design of the neural networks based on evolutionary computation is
present. A linear chromosome representation of the network are used by genetic
operators, which allow the evolution of the architecture and weights
simultaneously without the need of local weights optimization. This paper
describes the approach, the operators and reports results of the application of
this technique to several binary classification problems.