From: QXUXBTVOTSAO@spammotel.com   
      
   "the 'dude'"  wrote in message news:<40717   
   fe$1@news.unimelb.edu.au>...   
   > >   
   > Right now I don't know what I should do.....experiment with larger mutation   
   > rates to keep diversity and prevent premature convergence into local   
   > minimums..... I have heard that small populations with larger mutation rates   
   > can be more efficient than larger populations with smaller mutation rates.   
   > Shit- I worry that my endeavor is not feasible and that my search space is   
   > too damn big.   
      
   I have two somewhat radical ideas you may wish to try:   
      
   1)   Wrightian islands with step-wise rejoining.   Divide your large   
   population into many small sub-populations, and run them with a high   
   mutation rate.   What you hope to get from this is that on some   
   island, the population may stumble onto a non-local peak.   But if   
   they indeed find a better solution to some part of the problem, they   
   are likely to have degenerated in many other ways.  It is thus   
   difficult to recognize success.   So rather than moving forward by   
   choosing the island with the most fit population, recombine the   
   islands pairwise.  (so if you ran 16 islands, join 1 with 2, 2 with 3   
   - until you have 8 larger islands)   Simulate for a while.   Recombine   
   pairwise again, and simulate, and so on until the whole population has   
   been rejoined.     A novel good solution to some part of the problem   
   has a chance to bubble up, so it ends up prevailing in the whole   
   population.    This requires that you mate the individuals - the   
   offspring get a random mix of the genes of their parents.   
      
   Run with a high enough mutation rate, and small enough islands, that   
   fitness declines on almost every island.   It will be in the rejoining   
   phase that the fitness gains show up.   I have tried this only with   
   artificial fitness functions.   It worked like magic.   But I'd be   
   very interested to know if it works with a real-world problem.   
      
   2)   Sex.   If the individuals mate sexually, that means that each   
   simulated individual has two genes - perhaps the same and perhaps   
   different - for each locus.   You have to decide how you want the   
   individual organism to respond when it has two genes - that is two   
   different rules - concerning the same situation.   
      
   Good luck.   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)