Evolutionary dynamics of structured genetic algorithms

For now I just copied the text of the main project page, I will add more tomorrow (Felix):

Modeling evolutionary dynamics of spatially structured populations -- I am very interested in the interplay between ecology and evolution. Motivated by Sewall Wright's seminal 1932 paper I would like to investigate the effect that metapopulation formation has on the speed of evolution. The computational model to investigate this could be based on work by Mitchell & Crutchfield (and coworkers) where a genetic algorithm evolves a population of cellular automata to perform a certain task (see this paper). I propose to embed population structure in the genetic algorithm (GA) and find out what effect this has on the capabilities of the GA. One way of doing this might be to (bluntly) define subpopulations that cross breed at specified intervals; but I am sure that there are much more elegant/sophisticated ways of embedding structure. Some (wild) ideas for this are: putting them on graphs or lattices (with or without vacancies/movement)…. Any suggestions are greatly appreciated! (also other ideas for modeling stuff like this (evolutionary game theory etc..) are welcome)

• I have the same idea and I am thinking about how to train the initial structure to be the one robust to different situations. (Xin Wang)
• What about a metapopulation approach on a cellular automata? Each cell is a metapopulation, therefore giving local rules for the metapopulation and then global rules for the entire population (including migration). (Megan Olsen)
• I was considering introducing population structure in my artificial evolution system (which uses a GA now). It is not based on cellular automata but on artificial gene regulatory networks. (Borys Wrobel)
• (Andrew Hein) I have some ideas based on metapopulation and metacommunity models that may be pertinent. Sounds like a cool problem.
• Many ideas from population genetics suggest that metapopulations may be able to 'solve' problems that single populations could not, it will be interesting to see how this impacts genetic algorithms...especially as it pertains to robustness (Samuel_Scarpino)
• lets meet to chat about this project tonight (tuesday 8/6) at 8:30
• some relevant papers are: [1] and [2]