Docstoc

Genetic Algorithms

Document Sample
Genetic Algorithms Powered By Docstoc
					Genetic Algorithms




                     Sanchit Arora
                     2005CS10182
                     Group 1
Outline
   Introduction
   Inspiration
   History
   Setup and Methodology
   Strengths and Limitations
   Application Domains
Introduction
 Programming technique
   Mimics biological evolution
     Problem solving strategy
 Adaptive Heuristic Search
  Algorithm
 Particular class of evolutionary
  algorithms
Inspiration
 Nature: Evolution
 Survival of the fittest
 Evolutionary Biology
     Selection
     Inheritance
     Mutation
     Crossover (recombination)
History
 Model natural evolution(50‟s-60‟s)
 Machine learning techniques „62
 Ingo Rechenberg „65
   Evolution strategy: more similar to hill climbing
 Evolutionary programming „66
   Finite state machines->mutations
 John Holland (60‟s-70‟s)
   Proposed crossover and other recombination
    operators
Setup
 Requirements for a typical genetic
  algorithm
   A genetic representation of the solution
    domain
     Binary encoding
     Arrays of integers
     Strings of letters
   A fitness function to evaluate the
    solution domain
     Problem dependent
Methodology
 Initialization
   Random initial population
   “Seeded” initial population
Methodology contd…
 Selection
     Elitist selection
     Fitness proportionate selection
     Roulette-wheel selection
     Scaling selection
     Tournament selection
     Rank selection
     Generational selection
     Steady-state selection
     Hierarchical selection
Methodology contd…
 Reproduction
   Genetic operations
     Crossover
       Single point crossover
       Uniform crossover
     Mutation
   Expected increase in average fitness
Methodology contd…
 Termination
     Solution found
     Max number of generations reached
     Resources saturated
     No progress
Pseudo-code
 Choose initial population
 Evaluate fitness of each individual in
  population
 Repeat until termination:
   Select best ranking individuals to reproduce
   Breed new generation through genetic
    operations
   Evaluate individual fitness of each offspring
   Replace worst ranked part of population with
    offspring
Strengths
 Intrinsically parallel
 Well suited to problems with large solution
  space
 Works even for complex fitness landscapes:
     Noisy
     Discontinuous
     Changes over time
     Many local optima
 Ability to manipulate many parameters
  simultaneously
 Open-minded approach
   Negative feedback
Limitations
 Defining problem representation
 Writing fitness function
 Other parameters
   Size of population
   Rate of mutation and crossover
   Type and strength of selection
 Deceptive fitness functions
 Premature convergence
 Unnecessary for analytically solvable
  problems
Application Domains
   Electrical engineering
   Financial markets
   Game playing
   Mathematics and algorithms
   Molecular biology
   Pattern recognition and data mining
   Robotics
   Routing and Scheduling
   Systems Engineering
References
 Wikipedia
 The TalkOrigins Archive
   http://www.talkorigins.org/faqs/genalg/genalg.html

 Ai-junkie
   http://www.ai-junkie.com/ga/intro/gat1.html

 Imperial college documents
   http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol
    4/tcw2/report.html.26171
Thank You

  “The random chance of variation, coupled with the law of
  selection, is a problem-solving technique of immense
  power and nearly unlimited application”
                   :an insight on results by Charles Darwin