(Intro To) Evolutionary Computation Revision Lecture Ata Kaban The University of Birmingham Overview • • • • Overview of key notions and techniques Example questions Revise worked problem solutions Taking questions Representation • Deciding on the representation is the first step in designing an EA application • We had examples of – – – – Binary Real valued Trees (GP) Special, e.g. • Order-based: in the TSP problem, need to repr tours • Rule-based: need to represent sets of rules • Representation for NNs Q: Could you decide on a suitable representation when given a problem description? Genetic Operators • Depend on the representation • Mutation-type (one parent) • Crossover-type (typically two parents) • Self-adaptation Q: Can you describe crossover and mutation operators for each representation scheme? Q: Can you describe differences between different crossover or mutation operators? Q: Can you say when, how and why would you use self-adaptation? Fitness Computation and Selection Schemes • Selection schemes – Roulette, tournament, ranking, … • Fitness Sharing, Niching, Crowding – These are methods to control population diversity • Q: Could you list advantages and disadvantages of different selection schemes • Q: Could you explain the differences between explicit fitness sharing and implicit fitness sharing as well as their advantages and disadvantages? Other topics • Co-Evolution – Competitive or cooperative – One or several populations • Constraint Handling – Penalty approach (static, dynamic, adaptive) – Repair approach – Others (by co-evolution, by multi-obj, by designing specialised operators that preserve the constraints) • Multi-objective Optimisation – Pareto-optimal solution Revise Example Problems • We gave loads of examples all over the place in the lecture to illustrate notions or techniques. We have also worked through detailed solutions to some – very important to revise them! – Function optimisation – Co-evolution: Iterated Prisoner’s Dilemma – Combinatorial optimisation: Travelling Salesman Problem – Classifier systems & evolving NN – e.g. could you devise a solution to weather prediction? Types of questions • A few easy general technical questions • Specific technical questions • Problem solving questions: given a problem description (close to those we had), design an appropriate EA solution # No question requires you to know formulas! # You can use textual explanation, figures, pseudocode, formulas or whatever is more comfortable for you to express your answer. Don't forget to revise the last few lectures' topics either! - Estimation of Distributions Algorithms (EDA) - Theory of EA Some more advices • Make sure you know where the exam takes place • Even if you don’t know the complete answer, write as much as you do know. – We give some points for partial answers also • Use examples to help you explain things • Cover as many questions as you can – Don’t spend all your time giving a brilliant answer for one question only as there is a limited number of points we give for each question • Think a bit before you answer Good Luck!