Traps and Dangers when Modelling Problems for Genetic Algorithms
Stefan Wagner, Michael Affenzeller, Daniel Schragl
Abstract:
This paper describes the relevant steps when modelling a given problem in order to be attacked by a Genetic Algorithm (GA). As an example for the whole modelling process the cryptanalysis of a generic 3-rotor machine is used. Thereby the authors want to especially highlight the various traps and dangers that might lead to a complete failure of a GA-based approach. Finally a theoretical analysis is given that shows why problems like the used example cannot be solved by Genetic Algorithms sufficiently.
Bibliographical:
@inproceedings{ wagner04a, author = "S. Wagner and M. Affenzeller and D. Schragl", title = "Traps and Dangers when Modelling Problems for Genetic Algorithms", booktitle = "Cybernetics and Systems 2004", editor = "R. Trappl", publisher = "Austrian Society for Cybernetic Studies", volume = "1", pages = "79--84", year = "2004", url = "citeseer.ist.psu.edu/wagner04traps.html" }
URL:
http://citeseer.ist.psu.edu/635599.html
Notes:
Interesting overview of what it takes to develop GA systems. I disagree with some of the points (I don't think a representation needs to be simple and intuative. It needs to be good for evolving.) Pretty high-level, but still raises some important issues, such as the relationship between the fitness function and the correctness of the solution.