CSCI 6506 - Genetic Algortihms and Programming

Final Project

• Basically takes the form of a 5 page (max) 2 column IEEE formated paper due at the last lecture of term.
• List of Example projects.

General Course Information

• Course Description
  
• Short GA/ GP Bibliography
  
• Web Resources
  

Assignments

• Assignment 1: Generic ML Concepts - Due 25 Sept (midnight; email mheywood@cs.dal.ca)
• Assignment 2: GP sand pit! -
  • Koza's Canonical Tree structured GP
  • Local copy of lilgp distribution
  • Thyroid data set and default interface for lilgp (Part 2 of assignment)
  • Some basic comments regarding the additional lilgp interface.
  October 11 (in class or email submission by midnight)

Group Presentations

• The presentation ``policy''
• Presentation Groups
• Presentation Material
  • Group 1: Symbiotic GA, Presenting Oct 16
  • Group 2: Linkage learning and the Extended Compact GA, Presenting Oct 23
  • Group 3: Strength Pareto Evolutionary Algorithm, Presenting Oct 30
  • Group 4: GP and Generalization, Presenting Nov 1
  • Group 5: On NeuroEvolution..., Presenting Nov 6
  • Group 6: Evolving keepaway soccer players through task decomposition, Presenting Nov 20
  • Group 7: Evolving Java bytecode, Presenting Nov 22

Course Presentation Slides

• Machine Learning `101' - Basic ML context (#1)
  
• Simple GA - Goldberg's simple GA by hand (#2)
  
• Job-Shop Scheduling - Significance of Representation (#3)
  
• Subspace clustering - Modularity and Symbiosis (#3)
  
• GP intro - GP and constructing models (#5)
  
• GP detector for Intrustion Detection System on KDD data set - Efficient training on half a million patterns (#9)
  
• Packet Routing - Insect metaphor verses distributed GA (#10)
  
• Crossover Biases in GP - Code Bloat and evolving parsimonious GP solutions (#11)
  
• Active Learning - Exemplar subsampling under GP (#11)
  
• Coevolutionary Multi-objective GP - Co-operative problem decomposition under supervised learning while scaling to large data sets(#12)
  
• Symbiosis - third form of coevolution. Also related to macro search operators and group fitness/ selection.
• Symbiotic Bid Based GP - hierarchical complexification and context learning in Reinforcement Learning.

Course Notes

• Terminology and Metaphors - Search Spaces (#1)
  
• Holland's Schema Theorem - Why should stochastic algorithms work! (#2)
• GA Application example - Hard C-means Clustering (#3)
  
• Species and Niching - Multimodal problem solving (#4)
  
• Deceptive Problems - Static Formulation and Significance of Deceptive Landscapes(#4)
  
• Grammatical Evolution - GA selection of CFG Production rules for automatic program design(#5)
  
  • Home page for Grammatical Evolution
• Generic solutions through recursion - Problem solving with recursion, without explicit support for iterative instructions(#6)
  
• Performance Metrics for GP - Computational Effort and Statistical metrics for qualifying GP performance(#6)
  
• GP schema thoery - Lower bound GP Schema Theorem for 1-point crossover(#7)
  
• Exact GP Schema Theory - Poli's work on exact schema theorems for 1-point crossover(#8)
  
• Empirical Evidence for GP Schema Theory - Empirical Evidence for GP Lower Bound Schema Theorem(week #8)
  

Course Assignments

• Assignment #1 - Generic machine learning and Evolutionary Computation
  
• Assignment #2 - GP sand pit
  
• Project milestone 1 - presentation: background, definition of outcomes and outline of approach;
• Project milestone 2 - basic implementation;
• Project milestone 3 - update of findings and recommendations;
• Project milestone 4 - presentation of final results and future work.
• Project milestone 5 - due date for final report.

Benchmark Dataset Repositories

• Concise Data Repository - collection of 46 different classification problems from Universal Problem Solvers Inc.
• UCI Machine Learning Repository - If you want the definitive article...