Neural Network Programming : With the incorporation of a genetic evolution algorithm
Pintér, Martin (2016)
Hämeen ammattikorkeakoulu
2016
All rights reserved
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2016060311760
https://urn.fi/URN:NBN:fi:amk-2016060311760
Tiivistelmä
The author's aim in this project was to develop a neural network unit with the incorporation of a genetic evolution algorithm, experimenting with possibilities and the artificial recreation of neurological registration, using the programming language of C++.
The neural network evaluated input data into output data in a form to provide artificially intelligent response over different scenarios. This neural network was aided by the genetic evolution algorithm to distinguish bad results from good ones.
The author designed a simulation for the neural net to handle. This simulation tests solutions for an interception problem, running two rockets: a 'Bandit' which aims to shoot out its target and an 'Interceptor' which attempts to prevent this by eliminating the Bandit in time.
Many different scenarios were tested over different versions of the program, with the AI module reacting in different ways. The study showed that it is important to fitness calculation methods independent of expected mechanics.
The author implemented object orientated programming principles in order to make the code easily extendable for versatility and future development.
The project succeeded, the Interceptor has learnt to shoot the Bandit within different circumstances.
The neural network evaluated input data into output data in a form to provide artificially intelligent response over different scenarios. This neural network was aided by the genetic evolution algorithm to distinguish bad results from good ones.
The author designed a simulation for the neural net to handle. This simulation tests solutions for an interception problem, running two rockets: a 'Bandit' which aims to shoot out its target and an 'Interceptor' which attempts to prevent this by eliminating the Bandit in time.
Many different scenarios were tested over different versions of the program, with the AI module reacting in different ways. The study showed that it is important to fitness calculation methods independent of expected mechanics.
The author implemented object orientated programming principles in order to make the code easily extendable for versatility and future development.
The project succeeded, the Interceptor has learnt to shoot the Bandit within different circumstances.