Sale!
Placeholder

Evolution in a Neural Network: Finding a Light Source.

10,000 3,000

Topic Description

Summary
There are different ways to program an agent to navigate an environment.
These include hard coding every movement, logic based programs and inference engines built on top of lots of rules which ‘learn’ their way around the world. One interesting alternative is a neural network. This method allows for novel and flexible behaviours.
The main problem with setting up a neural network is finding the balance of nodes, thresholds and weights that define the complex architecture, such that the desired behaviour is achieved. Hand coding all of these variables is a task which is time consuming to say the least, if not impossible to achieve, and may not produce the most efficient outcome given more complex tasks.
One way to solve this problem is to use an evolutionary algorithm to evolve the architecture of the neural network. An evolutionary algorithm takes a population of agents and allows the more successful agents to survive, combining them with other agents in order that we produce a successful architecture which will succeed at the task at hand with maximum efficiency. Over time we should see
convergence to a solution. With this in mind, this paper will deal with creating an agent that uses a neural network to navigate a simple simulated world to find a light source. A genetic algorithm will then be implemented to evolve this neural network to create an agent which is successful at the task.

Contents
Summary I
Contents II
Chapter 1 – The Problem
Introduction …………………………………………………………………………………………… 1
What are the Benefits of Evolution? …………………………………………………….. 1
The Process of a Genetic Algorithm ……………………………………………………… 3
Explaining Genetic A lgorithms & Neural Networks ………………………………. 4
Considering Different Types of Crossover ………………………………………………. 6
Chapter 2 – Designing a Genetic Algorithm
The Idea ………………………………………………………………………………………………… 7
The World ………………………………………………………………………………………………. 8
The Neural Network ……………………………………………………………………………… 10
The Evolutionary Algorithm …………………………………………………………………. 14
Chapter 3 – Evaluation
Measuring Success ………………………………………………………………………….…… 17
A Summary of Observed Evolving Behaviour ………………………………….…. 19
Exper imental Results ……………………………………………….………………………….. 23
Chapter 4 –Conclusions
Conclusions Drawn From the Model ……………………………………………………. 31
Possible Expansion of the Project ………………………………………………………… 33
Survival as its own Fitness Function ……………………………………………………. 34
Appendix A – Personal Reflections III
Appendix B – Program Guide and Notes IV
Bibliography

PROJECT SAMPLE/DEPARTMENTS

REVIEW OUR SERVICES

SEE FAQ