Another One Bites the Dot Teaching a Neural Network to Play Pac-Man Using Biologically Motivated Learning Techniques

Date of Award

2010

Document Type

Thesis

Degree Name

Bachelors

Department

Natural Sciences

First Advisor

Henckell, Karsten

Keywords

Neural Network, General Algorithm, Video Game

Area of Concentration

Computer Science

Abstract

A branch of Artificial Intelligence known as natural computing, which proposes learning algorithms based on naturally occurring processes such as natural selection and biological processes such as neural networks, has been shown to be capable of learning and generalization of complex problems. This thesis evaluates two such biologically motivated techniques � temporal difference learning, which learns by reinforcement, and a genetic algorithm, which learns by natural selection � by teaching a neural network to play the classic arcade game Pac-Man. A replica of Pac-Man was created using the Python programming language, and networks trained with temporal difference learning, a genetic algorithm, and a combination of the two, each for 32 hours, were shown to exhibit concrete strategies and a significant improvement compared to a random solution.

Rights

This bibliographic record is available under the Creative Commons CC0 public domain dedication. The New College of Florida, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.

This document is currently not available here.

Share

COinS