Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Emergent Evolution allows Meta Evolution Adam Nellis University of York Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Evolutionary Algorithms How many? How do they solve the problem? Initialise a population of individuals How? Assess fitness of individuals How many? Using what method? Select some individuals to survive How? Reproduce these individuals, with variation What variation? How much variation? Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Parts of an Evolutionary Algorithm Initialise a population of individuals Assess fitness of individuals Select some individuals to survive 1. A problem to solve 2. A population of candidate solutions to the problem 3. A way of improving the quality of solutions in the population Reproduce these individuals, with variation Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Evolution is not Evolutionary Algorithms • Evolution is just: – Variation, inheritance and selection • It happens because: – Individuals reproduce – Reproduction is not perfect – Resources are limited (so there is selection) • Over time, this makes fitter individuals – But what is an “individual”? Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Emergent Evolution Solutions to problems evolve Evolutionary Algorithm parameters Evolution emergence selection Organisms living, dying and reproducing evolve Evolutionary Algorithm parameters evolve ? produce Solutions to problems ? Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Evolutionary Algorithm in the phenotype Phenotype Fitness evaluate Phenotype inform Evolutionary Algorithm develop Genotype Individual Problem-solving machinery Expression machinery Copying machinery change express Genotype copy Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 How? Individuals that are alive! 1. Perform a task (solve the application problem) 2. Metabolise food to gain energy 3. Use energy to express genes to do points 1 and 2 4. Copy my genes (imperfectly), to reproduce 5. Encode all of the above functionality in the genes Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Example – Stringmol A self-copying machine Evolving over time Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Large-scale mutation 09: OBEQBXUUUDYGRHBBOSEOLHHHRLUEUOBLROORE$BLUBOˆB>C$=?>$$BLUBO%}OYHOB 31: BBOSEOLHHHRLUEUOBLROORE$BLUBOˆB>C$=?>$$BLUBO%}OYHOB 09: OBEQBX...UOBLROORE$BLUBOˆB>C$=?>$$BLUBO%}OYHOB 29: OBEQBX...UOBLROORE$BLUBPˆB>C$=?>$$BLUBO%}OYHOB 30: OBEQBXUUUDYGRHBBOSEOLHHHRLUEUOBLR OORE$BLUBPˆB>C$=?>$$BLUBO%}OYHO OBEQBXUUUDYGRHBBOSEOLHHHRLUEUOBLR OORE$BLUBOˆB>C$=?>$$BLUBO%}OYHOB Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010 Summary • Meta evolution via emergent evolution – Not by evolving parameters in a hardcoded algorithm • Richer individuals – That have the Evolutionary Algorithm as part of their phenotype • Example of large-scale-mutation – Strings copying each other with per-character mutation – Evolved a large-scale macro-mutation Emergent Evolution allows Meta Evolution – Adam Nellis – EmergeNet 4: Engineering Emergence – 19th April 2010