carasso@cory.Berkeley.EDU (George of the Jungle) (10/24/88)
I was interested in designing an experimental program, where the program would try to solve a problem by using various heuristics. What I would like to do is model it after evolution. If a heuristic is successful, then it creates mutant heuristic versions of itself, and then those heuristics are put to the test, et cetera. Each time, the ones that solve the problem in the shortest time are allowed to have children, others are "terminated". In wanting mutant children created, I do not simply want to change some constants in the parent's fuctions, but rather would like it to change the function itself and create totally new ones. In starting this off, I also do not want to enter the base case for all the fuctions I know. For example \x. x*x; or \x. c; or \x. x^x; or what ever fuction I could dream up. Is there any way to create "all function groups", within reason? Where two functions are in the same group if they differ by a constant, or number of repeatitions. Of course, the eventual goal is to see the program find a good heuristic on its own. Roger Carasso, UCB "My ignorance is my own, and is no way related to any organization"
thefool@athena.mit.edu (Michael A. de la Maza) (10/25/88)
<I was interested in designing an experimental program, where the program <would try to solve a problem by using various heuristics. What I would like <to do is model it after evolution. <Roger Carasso People involved in writing genetic learning algorithms are involved in projects much like the one you describe. To learn more about glas I suggest that you join the gla mailing list moderated by John Grefenstette <GA-List-Request@AIC.NRL.NAVY.MIL>. -- Michael de la Maza thefool@athena.mit.edu
nau@mimsy.UUCP (Dana S. Nau) (10/25/88)
In article <6746@pasteur.Berkeley.EDU> carasso@cory.Berkeley.EDU.UUCP (George of the Jungle) writes: >I was interested in designing an experimental program, where the program >would try to solve a problem by using various heuristics. ... If a heuristic >is successful, then it creates mutant heuristic versions of itself, and then >those heuristics are put to the test, et cetera. ... Ping-Chung Chi has done a nice Ph.D. dissertation at the University of Maryland, studying game tree searching. Among other things, he has done the above on game trees using a genetic algorithms approach. For more information, write to chi@mimsy.umd.edu. -- Dana S. Nau Computer Science Dept. ARPA & CSNet: nau@mimsy.umd.edu University of Maryland UUCP: ...!{allegra,uunet}!mimsy!nau College Park, MD 20742 Telephone: (301) 454-7932