Searching for optimal size neural networks

Assembler Encoding represents a neural network in the form of a simple program called Assembler Encoding Program. The task of the program is to create the so-called Network Definition Matrix, which maintains all the information necessary to construct a network. To generate the programs and, in consequence, neural networks, evolutionary techniques are used. One of the problems in Assembler Encoding is to determine an optimal number of neurons in a neural network. To deal with this problem a current version of Assembler Encoding uses a solution that is time consuming and hence rather impractical. The paper proposes four other solutions to the problem mentioned. To test them, experiments in a predator-prey problem were carried out. The results of the experiments are included at the end of the paper.