The Influence of Fitness Function on Quality of Neuro-Controllers Produced with Assembler Encoding

摘要

Assembler Encoding is a neuro-evolutionary method in which neural networks are represented in the form of evolutionary created programs. Each program consists of operations and data evolving in many separate populations. To enable the evolution to produce effective programs and in consequence neural networks, operations and data from each population have to be reliably evaluated. The paper presents and compares several methods that can be used for that purpose. The comparative tests were carried out on the predator-prey problem.