artificial intelligence in information processing genetic algorithms by theresa kriese for...

26
Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Post on 22-Dec-2015

228 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Artificial Intelligence in Information Processing

Genetic Algorithms

by Theresa Kriesefor Distributed Data Processing

Page 2: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Content• Introduction

• Understanding: Travelling Salesman Problem

• Biological Background

• GA’s in Information Processing

• Summary

• Sources

Page 3: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

How to solve problems, that are socomplex, that you can not get an exact

solution in an appropriate time?

Page 4: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Travelling Salesman Problem

The Problem: A Salesman needs to go to n citiesfor work. In each city, he has one customer.Because he doesn’t want to travel so long, heneeds to find the shortest possible route. He knowsthe single distances between two cities.

optimisation problem

not one solution, but the best possible no wrong or right solution

Page 5: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

How could you solve the problem?

• for all possible ways, the distance must be found

• with increasing n, the problem soon gets too complex

NP-Algorithm: problem can’t be solved in polynomial time / the needed calculating steps can’t be described by a polynomial

that’s more, than the amount of elementary particles in the universe!

* 10 cities = over 180000 possibilities* 24 cities = 1.3*10^22 poss.* 120 cities = 6*101^96 poss.

Page 6: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

This can’t be the right way..

no combinatorial solving

• For practical use:Instead of an optimum

(shortest route ever) after a long time

It’s better to get a suboptimum (short, but probably not the shortest) in the short-run

Example: optimal route for visiting the 15biggest cities in Germany

Page 7: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Let’s ask the

nature! She is

solving complex

problems for

hundreds of

centuries!

Page 8: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Biological Background

Different processes

during the reproduction

of a population in a long

period of time aspire a

perfectly adapted group

of individuals in the end.

Image: http://softwarecreation.org/images/2008/natural-selection.png

Page 9: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Mutation

Images: http://neatorama.cachefly.net/images/2006-07/albino-squirrel-white.jpg, http://employees.csbsju.edu/HJAKUBOWSKI/classes/SrSemMedEthics/Human%20Genome%20Project/mutation2.gif

Page 10: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Selection

Image: http://www.scienceteacherprogram.org/biology/NaturalSelectionIllustration.gif

Page 11: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Recombination (Crossover)

Image: http://en.wikipedia.org/wiki/Image:Morgan_crossover_1.jpg

Page 12: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

But how can we use it in Information Processing?

Page 13: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

When do we need genetic algorithms?

• Timetabling problems

• Bioinformatics

• Code-breaking

• Software Engineering

• Scheduling applications

• Marketing analysis

• File allocation for distribution systems

• Learning algorithms in neural networks

Page 14: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

How does it work?

-Different solution candidates

- fitness function

-Selection

-Mutation

-Recombination

-if break-up criteria is fulfilled

Best found solution

Page 15: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Steps in practice

• Initialisation

- generation of all possible “individuals” (solution candidates) by chance 1st generation

- encoding to binary code

• Evaluation

- using a fitness function, the fitness of each solution candidate is calculated

Page 16: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Process • Selection

- random selection of solution candidates

- the higher the fitness, the higher the probability to be selected

• Mutation

- random modification of candidates

Page 17: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

• Recombination (crossing over)

Mutation and crossing over are methods to generate a 2nd generation population.

New generation replaces worst ranked parts of the generation before.

Due to the repeating processes, the generations are getting closer to an optimum.

The whole process continues until a break-up criteria occurs.

Page 18: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Example

Page 19: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing
Page 20: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing
Page 21: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Images: http://fbim.fh-regensburg.de/~saj39122/vhb/NN-Script/script/gen/k040401.html

Page 22: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Understanding:Scheduling

• Example: hospital

• working in shifts

• many factors to consider: - law regulations - personal wishes for days off - shift premium - certain amount of doctors and nurses

• very complex information cluster in one big database

• program works out optimum schedule by using genetic algorithms

Page 23: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing
Page 24: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Summary

• Based on the biological evolutionGenetic operators used: - selection - mutation - recombination

• Developed to solve optimisation problems

• Can not give an exact solution but is approaching an optimum

Page 25: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

Sources

• www.wikipedia.org [en,de]

• www-e.uni-magdeburg.de/harbich/genetische_algorithmen [de]

• www.htw-dresden.de/~iwe/Belege/Boerner/ [de]

• http://www.uni-kl.de/AG-AvenhausMadlener/tsp-ger.html [de]

• http://www.sciencedirect.com Volume 39, Issue 5, September 2003, Pages 669-687 [en]

• www.fbim.fh-regensburg.de [de]

Page 26: Artificial Intelligence in Information Processing Genetic Algorithms by Theresa Kriese for Distributed Data Processing

THANK YOU FOR ATTENTION!