Genetic algorithms
-
Upload
mathijs-van-meerkerk -
Category
Education
-
view
1.343 -
download
7
description
Transcript of Genetic algorithms
![Page 1: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/1.jpg)
Genetic Algorithms
![Page 2: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/2.jpg)
What are we going to talk about?
• Evolution in biology• Evolution in programming• Components of a genetic algorithm• Strengths and limitations• Real world examples
![Page 3: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/3.jpg)
Evolution in biology• Darwin
• Variation
• Selection
• Heredity
![Page 4: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/4.jpg)
Evolution in biology• Modern evolution theory
• Discovery of DNA
• The gene as heredity and variation operator
![Page 5: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/5.jpg)
Evolution in programming
![Page 6: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/6.jpg)
Evolution in programming
![Page 8: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/8.jpg)
Component - Representation
Phenotype Genotype
Frame: 16 parameters
3 wheels: 9 parameters
“DNA” string of 25 values
8.7 4.7 9.5 5.9 6.2 1.3 … 9.3 1.0 8.7 3.8 1.7
![Page 9: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/9.jpg)
Component - Population
![Page 10: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/10.jpg)
Component - Fitness Function
Distance traveled
![Page 11: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/11.jpg)
Component - Parent selection mechanism
0,1 0,5
0,4 0,1
Chance forreproduction
![Page 12: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/12.jpg)
Component - Variation operators
part of parent 1 + part of parent 2 + mutation
8.7 4.7 9.5 5.9 6.2 1.3 1.8 … 1.0 8.7 3.8 1.7
8.5 9.1 0.1 3.5 8.0 1.5 2.2 … 1.6 5.5 2.8 4.5
8.7 4.7 3.4 5.9 6.2 1.7 2.2 … 1.6 5.5 2.8 4.5
Parent 1Parent 2
Offspring(+ mutation)
![Page 13: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/13.jpg)
Component - survivor selection mechanism
![Page 14: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/14.jpg)
Strengths• “Parachutist” comparison
– Parallelism:• multiple parachutists
– Mutation:• search in vicinity
– Selection:• focus on most successful
parachutists– Reproduction:
• give them walkie-talkies
![Page 15: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/15.jpg)
Strengths
• Parallelism– CPU efficient– Problems with multiple objectives
• Blind watchmakers– Randomness = open mind
• Anytime behaviour
![Page 16: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/16.jpg)
Strengths
• Wide range of applications
![Page 17: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/17.jpg)
Limitations
• Defining the problem / strategy– Genotypes– Fitness function– Rate of mutation / crossover– Parental selection
• Premature convergence• Analytically solvable problems
![Page 18: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/18.jpg)
Examples - Anaconda
• Kumar Chellapilla David B. Fogel
• Expert level
![Page 19: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/19.jpg)
Examples - Anaconda
• your checker +1, enemy checker –1, no checker 0
• doesn’t know winning condition
• 15 neural networks that competed against each other
![Page 20: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/20.jpg)
Examples - Faceprint
• Make composition drawings of criminals
• Hard to describe individual facial features
![Page 21: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/21.jpg)
Examples - Faceprint
• 5 numbers for facial features
• 5 proportion
• Fitness is how close the picture is to the criminal
![Page 22: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/22.jpg)
Examples – Movie Recommenders
• Film database– Genres
• How much will you like a filmif it is from the X, Y and Z genres?
• Neural network– Trained with Evolutionary Algorithm
![Page 23: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/23.jpg)
Examples – Movie Recommenders
Score(will you like it?)
Genre (0 or 1):
Action
Adventure
Animation
Biography
…
Western
![Page 24: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/24.jpg)
Examples – Movie Recommenders
• Phenotypes: 240 weights
• Genotypes: 8-bit floats
• DNA = long binary string– E.g. 110010010010100011 … 00100001010100
• Population of 200 DNA-strings– Evolved by crossover, mutation & selection
![Page 25: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/25.jpg)
Examples – Movie Recommenders
• 11% better predictor than before
![Page 26: Genetic algorithms](https://reader033.fdocuments.us/reader033/viewer/2022061223/54c4947a4a795917618b4720/html5/thumbnails/26.jpg)
Thanks
Are there any questions?