Hello, all! I''ve been reading up on Evolutionary Programming (Koze, Angeline, Pollack, Fogel) and am absoulutely fascinated. I was actually considering such a system to evolve ghost control in a Pac-Man game. Does anyone have suggestions of readings that might be helpful? Examples would be great, and internet searchs don''t turn up much. Other question: I see a number of folks talking about their Master''s degrees. Where are people getting such training in AI? -Kirk

Evolutionary programming is to do with specifically evolving programs from programming primitives (I think they do it in LISP or whatever). Personally I''ve only ever used genetic algorithms, which tend to be applied to other paradigms such as neural networks or heuristics (probably what you''re looking for). Either way, most of my sources have been directly from lectures, which you can find here http://www.cogs.susx.ac.uk/users/inmanh/easy/alife/index.html or from scientific research papers. The lecture notes cover many aspects of artificial life in general and are well worth a read. An excellent introduction to the topic. Obviously they''re not indexed and just in order of our lectures.

Also Ferretman has a section on genetic algorithms here http://www.gameai.com/genetics.html which are more likely to be related to game AI.

I''ve used genetic algorithms constantly for the last six months so I know a few downfalls and ways to use them if you have a specific question.

My training is at Sussex University in the cognitive science department on a course titled "Evolutionary and Adaptive Systems". Personally I love it and it''s just got an increase in funded places so it''s worth applying if you have a 2:1 or better.

Mike

Mike,

Thanks for the tips and the links! As for the degree, I''m in the US, so that''s a limitation. I''m taking a job in Princeton, NJ soon, so I might be able to work toward a MS at either Rutger''s or Princeton. Not bad options.

As for the GA, yes a neural net was one idea. Another was decision control evolution. Such a system would likely evolve complex if..then structures with varying depths. The hope is to establish one GA for the ghosts and one for the pac-man, then play them in a tournament fashion and see what falls out. Kind of a co-evolutionary strategy a-la Pollack, Angeline, and Fogel. The neural net sounds interesting too, though. Ah, so many toys so little time to play with them all.

Thanks again for the info!!

-Kirk

Genetic programming is really simple. A really simple implementation is by "mating" 2 properties of two together. For example it player 1 has 20% dexterity, and player 2 has 50% and they mate. Then you get a random number in between. Or if it is a mutant it can be something larger. That''s the basic idea. www.geocities.com/wavewash that''s a program I did a year back. It uses this simple idea. Hope it helps.
~Wave

In the Netherlands there is a normal study called Kunstmatige Intelligentie (dutch translation of AI). You can follow this at many universities. Right now, I am going to the University of Amsterdam...

quote: Original post by Wavewash

Genetic programming is really simple. A really simple implementation is by "mating" 2 properties of two together. For example it player 1 has 20% dexterity, and player 2 has 50% and they mate. Then you get a random number in between. Or if it is a mutant it can be something larger. That's the basic idea. www.geocities.com/wavewash that's a program I did a year back. It uses this simple idea. Hope it helps.
~Wave

Just so there is no confusion for other readers, what you are describing is more akin to using a Genetic Algorithm as opposed to Genetic Programming (as described by Koza, Fugel, et al).

Eric


Edited by - Geta on May 14, 2001 10:48:06 AM