Genetic Algorithms
6 likes2,688 views
The document introduces genetic algorithms, which are inspired by biological evolution. It describes how genetic algorithms use operations like selection, crossover and mutation to evolve solutions to problems in a way that is analogous to natural selection. It also outlines the basic components of a genetic algorithm, including representing solutions, initializing a population, evaluating fitness, and selecting solutions to breed new generations. Finally, it discusses some common applications of genetic algorithms to optimization problems.
Genetic Algorithms
- 1. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions GENETIC ALGORITHMS Muhammad Adil Raja Roaming Researchers, Inc. August 12, 2014 Muhammad Adil Raja Genetic Algorithms
- 2. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 3. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 4. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 5. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 6. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 7. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 8. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 9. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions OUTLINE INTRODUCTION BIOLOGICAL INSPIRATION SEARCH SPACES A GENETIC ALGORITHM EXPERIMENTAL SETUP GENETIC OPERATORS APPLICATIONS CONCLUSIONS Muhammad Adil Raja Genetic Algorithms
- 10. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions INTRODUCTION TO GENETIC ALGORITHMS (GAS) Genetic algorithms are inspired by Charles Darwin’s theory of evolution. Fall under the umbrella of evolutionary computing. Idea came from John Holland. Muhammad Adil Raja Genetic Algorithms
- 11. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions BIOLOGICAL INSPIRATION I The idea is inspired from natural evolutionary biological systems. In natural biological evolutionary systems, organisms are made of cells. A cell is composed of a set of chromosomes. Chromosomes are found in the nucleus. Chromosomes are made of DNA. FIGURE: Structure of a Biological Cell Muhammad Adil Raja Genetic Algorithms
- 12. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions BIOLOGICAL INSPIRATION II Sections of Chromosomes are called genes. DNA - deoxyribonucleic acid. it is the genetic code that contains all the information needed to build and maintain an organism. FIGURE: Chromosome Structure Muhammad Adil Raja Genetic Algorithms
- 13. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions BIOLOGICAL INSPIRATION III Each organism has a distinct number of chromosomes. In humans every cell contains 46 chromosomes (23 pairs). Other organisms have different numbers. A dog has 76 chromosomes per cell. Chromosomes come in pairs. These are called homologous pairs (homologs). Homologs can be imagined as matching pairs. But they are not exactly alike. Like a pair of shoes they can be different. Muhammad Adil Raja Genetic Algorithms
- 14. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SOME JARGON I Chromosomes are composed of DNA. DNA (and consequently chromosomes) are made of genes. A chromosome contains hundreds of thousands of genes. Trait: Each gene encodes a particular protein, e.g. eye color. Alleles: Possible settings for a trait (e.g. color can be blue, brown or black). Muhammad Adil Raja Genetic Algorithms
- 15. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SOME JARGON II Locus: Each gene’s own position in chromosome. Genome: Complete set of genetic material. Genotype: A particular set of genes in a genome. Phenotype: A genotype’s physical and apparent characteristics. (e.g. color, height, intelligence etc.) Muhammad Adil Raja Genetic Algorithms
- 16. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions REPRODUCTION I Crossover: (recombination): Happens during reproduction. Genes from parents recombine in a meaningful sense to form a whole new chromosome. Offspring. They can be genetically mutated. Muhammad Adil Raja Genetic Algorithms
- 17. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions REPRODUCTION II Mutation: Elements of the DNA are randomly changed a little bit. This change is mainly caused during reproduction by errors committed during copying genes from parents. Fitness: A measure of success of the organism in a typical ecosystem. Muhammad Adil Raja Genetic Algorithms
- 18. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SEARCH SPACES I Space of all feasible solutions. Each point in a search space represents one feasible solution. Each feasible solution can be marked by its value or fitness for a problem. Good solutions are desired. It is often not possible to prove what is an optimum solution. FIGURE: A Non-Linear Search Space Muhammad Adil Raja Genetic Algorithms
- 19. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SEARCH SPACES II Search spaces can be very non-linear. Like a mountainous terrain. Finding the optimum solution is the real challenge. Many locally optimum solutions can exist. One or few globally optimum solutions may also exist. How to find the best one? That is what optimization is all about. Muhammad Adil Raja Genetic Algorithms
- 20. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions A GENETIC ALGORITHM I Solutions to problems are actually evolved. The algorithm starts with a set of randomly chosen solutions. The solutions can be good or really really bad. Solutions are evaluated for their fitness. Muhammad Adil Raja Genetic Algorithms
- 21. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions A GENETIC ALGORITHM II Solutions from one population are taken and used to form a new population of better solutions. Solutions that are selected to form new offspring solutions are selected according to their fitness. The more suitable ones have more chances to reproduce. The algorithm is repeated until some stopping criterion is met. Muhammad Adil Raja Genetic Algorithms
- 22. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions A GA LIFE CYCLE: THE PSEUDOCODE A GA Life Cycle: The Pseudo Code 1. Create an initial population of candidate solutions to a given problem. 2. Evaluation. 3. Selection. 4. Reproduction. 5. Evaluation. 6. Replacement. 7. Continue from 3. Muhammad Adil Raja Genetic Algorithms
- 23. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions A TYPICAL GA BREEDING CYCLE Muhammad Adil Raja Genetic Algorithms
- 24. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions TABLE: Fiddle Parameters of a Typical GA Experiment Parameter Value Initial Population Size 300 Initial Tree Depth 6 Selection Tournament Selection & Roulette Wheel Tournament Size 2 Genetic Operators Crossover and Mutation Operators Probability Type Adaptive Initial Operator probabilities 0.5 each Survival Elitism and Generational Generation Gap 1 Muhammad Adil Raja Genetic Algorithms
- 25. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SELECTION Roulette Wheel Selection – Fitness Proportionate Selection. Tournament Selection. Muhammad Adil Raja Genetic Algorithms
- 26. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SOLUTION REPRESENTATION IN GAS I Depending upon the problem and its formulation, a solution can be represented in various ways in a GA. Most notable representations are: 1. Binary string representation. This is one of the most common way of representing a solution in a GA. The solution is represented as a string of binary numbers. Akin to a chromosome in biology. 2. Integer-valued arrays – Integer programming (?). 3. Real-valued arrays – for continuous parameter optimization. 4. Complete computer programs – as in GP. Muhammad Adil Raja Genetic Algorithms
- 27. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SOLUTION REPRESENTATION IN GAS II TABLE: Binary String Representation Chromosome 1 1101100100110110 Chromosome 2 1101111000011110 Muhammad Adil Raja Genetic Algorithms
- 28. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions CROSSOVER I 1. Randomly choose two individuals (chromosomes/individuals). 2. Choose crossover points on each one of them. 3. Swap the sub-parts around crossover points to form new offspring. Respect Syntactic or semantic constraints. The child should solve the problem somehow. Muhammad Adil Raja Genetic Algorithms
- 29. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions CROSSOVER II TABLE: Binary String Crossover Chromosome 1 110110010 0110110 Chromosome 2 110111100 0011110 Child Chromosome 1 110110010 110111100 Child Chromosome 2 0110110 0011110 Muhammad Adil Raja Genetic Algorithms
- 30. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions MUTATION I 1. Choose a newly created offspring. 2. Pick a random gene, or a few genes, on it. 3. Change its value to something else randomly – Change allele. Muhammad Adil Raja Genetic Algorithms
- 31. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions MUTATION II TABLE: Binary String Mutation Chromosome 1 1101100100110110 Chromosome 2 1101111010011110 Muhammad Adil Raja Genetic Algorithms
- 32. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions SURVIVAL Elitism Replacement Muhammad Adil Raja Genetic Algorithms
- 33. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions FITNESS EVALUATION Mean squared error (MSE). Chi squared error. Scaled mean squared error. Muhammad Adil Raja Genetic Algorithms
- 34. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions APPLICATIONS OF GAS I Applications are quite too many. GA as a hammer. A hammer that finds almost everything else as a nail. Muhammad Adil Raja Genetic Algorithms
- 35. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions APPLICATIONS OF GAS II In regression and classification. Regression or Classification of nonlinear problems. In Telecommunications: Speech quality estimation. In Computer Networks: Network coding. In Finance: In evolving effective bidding strategies. In Clinical: Cancer detectors, seizure detectors, mental health diagnosis etc. Muhammad Adil Raja Genetic Algorithms
- 36. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions APPLICATIONS OF GAS III In evolving chess players. In evolving antenna designs. Evolvable hardware. Muhammad Adil Raja Genetic Algorithms
- 37. Introduction Biological Inspiration Search Spaces A Genetic Algorithm Experimental Setup Genetic Operators Applications Conclusions CONCLUSIONS GAs are strong problem solving algorithms. They can be applied to a large number of optimization problems. Alternative solution representations render them suitable for a wide variety of problem domains. They are easy to understand. The analogue from biological evolution is quite helpful. They are easy to implement and fun to use. They can be used to solved difficult problems. Particularly suitable for finding acceptable solutions to otherwise intractable problems. ... Muhammad Adil Raja Genetic Algorithms