FUP: One thing about your demo intrigues me. After a number of genetartions (say 50 or so, depending on the number of hidden nodes) all your minesweepers are heading in generally the same direction. It struck me when I first saw it (almost like it is an emergent behaviour, which is quite rediculous since the sweepers have no notion of eachother whatsoever). Can you explain this?

Another thing that struck me pleasently: The little amount of nodes you can get by! The sweepers ran pretty wel with only 2 hidden nodes in 1 layer :-)

Sander Maréchal
[Lone Wolves GD][RoboBlast][Articles][GD Emporium][Webdesign][E-mail]

Because, over time, the sweepers'' genomes converge to be similar.


You can evolve sweepers that pick up mines efficiently with no hidden units at all...



ai-junkie.com

quote: Original post by kirkd
Timkin, are you a DeJong fan? 8^) I''ve seen a few of his papers and they''ve been pretty good by my standards, but I''m not aware of the reputation.

DeJong has been a long time researcher in GAs. He created some very good test suites of functions for evaluating the perfomance of GAs and for comparing their performance to other, more established algorithms. His work has always stood out in my opinion.

Timkin