First Things First This is 4003-590-02 / 4005-756-02 Welcome to - PDF document

First Things First  This is 4003-590-02 / 4005-756-02 Welcome to Genetic Algorithms (Genetic Algorithms)  I am Joe Geigel…your host! Plan for this afternoon First some thanks  Logistics  To Prof Butler for filling in for me.  What is this course about?  To WGBH for making that engrossing video on evolution.  Requirements and Deliverables  Time for attendance. Logistics Logistics  Course Web Site:  Course Web Site:  http://www.cs.rit.edu/~jmg/geneticAlgorithms  http://www.cs.rit.edu/~jmg/geneticAlgorithms  Contact:  Everything you need to know  office hours: MW 2-4 (or by appt)  Syllabus  Office: 70 (GCCIS) Rm 3527  Assignments  e-mail: jmg@cs.rit.edu  Schedules  phone: 475-2051  Reading List  Diary  Slides:  Will be available on Web site. 1

Logistics Logistics  Course Web Site:  Your RIT e-mail  http://www.cs.rit.edu/~jmg/geneticAlgorithms  Be sure that it is forwarded correctly.  Will be the authoritative source for the course.  myCourses  E.g. SCHEDULE has already been  Email changed.  dropboxes Logistics What is this course about?  Textbook  Evolutionary Computation by DeJong  Selected papers from GA literature (on Web site)  Supplement lectures  Not necessary, but nice to have as references. Top 5 misconceptions about this course. Evolutionary Algorithms Only cover GAs  An EA uses some mechanisms inspired by biological 1. evolution: reproduction, mutation, recombination, Will have an awfully hard final exam 2. natural selection and survival of the fittest. Hey, aren’t you Prof Anderson? 3.  Candidate solutions to the optimization problem play Will cover an awful lot of theory (leading to 4. the role of individuals in a population, and the cost an awful final exam) function determines the environment within which Long PBS documentaries will be shown each the solutions "live". 5. class.  Evolution of the population then takes place after the repeated application of the above operators. 2

Course objectives Getting back to the video  Upon completion of this course, students will  Things that you unexpectedly got from the be able to: video:  Explain the of the principles underlying  Monkeys are capable of learning the concept of 0 Evolutionary Computation in general and Genetic  Charles Darwin's brother is one of the best roles Algorithms in particular. I've ever seen on a PBS documentary.  Apply Evolutionary Computation Methods to find  Darwin married his cousin solutions to complex problems  Analyze and experiment with parameter choices in  finches can be blown really really far by the wind. the use of Evolutionary Computation  There were some really big rodents at one point  Summarize current research in Genetic Algorithms and Evolutionary Computing (GRAD Only) Thing I hoped you’d get from Thing I hoped you’d get from the video the video 1. Individuals evolve over generations.  2. Evolution is guided by fitness in a given environment. Thing I hoped you’d get from Thing I hoped you’d get from the video the video 3. Individuals are the “product” of their 4. An individuals traits can undergo parents. random mutation 3

Thing I hoped you’d get from the video Getting back to the video 6. Nature has no apparent 5. Genotype vs. phenotype goal (except to create individuals suitable to a given environment). Evolutionary Algorithms Evolution and Computing An EA uses some mechanisms inspired by biological evolution:  Your “world” is the context of a given  reproduction, mutation, recombination, natural selection and problem survival of the fittest. Candidate solutions to the optimization problem play the role of  individuals in a population, and the cost function determines the environment within which the solutions "live". Evolution of the population then takes place after the repeated  application of the above operators. Evolution Survival of the fittest  You will be evolving generations of  Evolution will be guided by fitness… “individuals”  “Individuals” are solutions to your  You must decide what makes a “fit” problem solution  Are you a good solution or a bad solution. 4

Genotype / Phenotype Solution breeding  Genotype (DNA) of your solution will be  Child solutions are spawned from a common data structure parent solutions  List, array, tree, etc.  Crossover  Mechanism to combine DNA to create DNA  Phenotype will be an “individual” of offspring solution derived from the DNA  You must define this translation. Solution mutation Solution mutation  The DNA from a single individual can  The DNA from a single individual can randomly mutate. randomly mutate.  Mutation  Mutation  Must define how given the data structure  Must define how given the data structure representing the genotype. representing the genotype. Goal Questions so far  To create individuals suitable to a given environment…  In other words…  Create “fit” individuals  Create “good” solutions  Create solutions that best solve the problem. 5

Evotionary Computation process Given a problem  To use evolutionary algorithms your must:  Define your genotype Initialize population  Identify your phenotype Select individuals for crossover (based on  Define the genotype -> phenotype translation fitness function Crossover  Define crossover and mutation operators  Define fitness Mutation  Determine selection criteria Insert new offspring  Set population parameters into population  This course will explore each part of the Are stopping criteria satisfied? process in depth! Finish Lectures Course deliverables  Lectures will cover:  In this course you will:  Process of using evolutionary algorithms  Choose a hard, multidimensional problem to solve.  Process applied to my favorite problem (TSP)  Define and code evolutionary methods to solve the problem.  Example uses of EAs in practice  Experiment with the parameters.  Listed as APPLICATION in schedule.  Guest lecturers.  You will be working in teams of 2. Teaming up Course deliverables  Four major deliverables:  Find a common problem…  Checkpoints / assignments  Implementation code  Mycourses interest questionaire:  Final report problem/area you would like to solve.  Presentation  Please fill in by Monday.  I will list students and topics on Web by  And for grad students: next class.  Grad report. 6

Checkpoints Report  Weekly (approx) assignments, each exploring  Final findings of how well your EC on task in using EC on your problem: solved your problem.  1. Framework  2. Problem Statement  Summary of findings in checkpoint  3. Genotype / Phenotype  4. Crossover / Mutation  5. Fitness / Selection  6. Population Parameters Code Presentation  Implementation of the EC:  Oral summary of your final report  From scratch  Last weeks of class + finals week.  Using a toolkit / API (some listed on Web  Peer evaluation. page)  Use language of your choice  Readable code please Grad report Grading  About grad work Graduate Undergrad  Help you with project / thesis Assignments 25% 30% Code 20% 25%  Report can be the background/previous Report 20% 25% work section of your grad report Presentation 15% 20%  Choose problem related to your work. Grad Report 20% 7

Ground rules Due dates  Use the language / toolkit of your  Checkpoints -- periodically (approx choice weekly)  Toolkit use is optional  Code + Report:  Work in teams of 2.  Presentations: week 9, 10, finals week  All submissions will be done  Grad report: finals week electronically using mycourses dropboxes. Questions  Next time:  EA Frameworks  Defining your problem  Please think about your problem and fill out mycourses questionairre. 8