The real problem connected with the construction of neural networks is to appoint the number of elements (neurons) and the connection structure. The algorithms, which construct the classification networks using ranked layers allow to establish the number of neurons and weights of their connections on the basis of training data set. In this paper, the method of ranked layer construction, using genetic algorithm has been presented. The applying the genetic algorithm leads to obtaining the network with less number of elements, hence the investigated network is of more ability to generalize information. Such a smaller network preserves its given classification property referring to the data of training set. In addition, applying of the genetic algorithm allows to use in the construction process any neuron activation function, e. g. rectangular activation function. This feature results in considerable reduction of the network size.
[1]
Kenneth Alan De Jong,et al.
An analysis of the behavior of a class of genetic adaptive systems.
,
1975
.
[2]
David E. Goldberg,et al.
Genetic Algorithms in Search Optimization and Machine Learning
,
1988
.
[3]
M. Golea,et al.
A Convergence Theorem for Sequential Learning in Two-Layer Perceptrons
,
1990
.
[4]
John H. Holland,et al.
Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology, Control, and Artificial Intelligence
,
1992
.
[5]
Zbigniew Michalewicz,et al.
Genetic Algorithms + Data Structures = Evolution Programs
,
1996,
Springer Berlin Heidelberg.