🚀 go-pugleaf

In article <Btx892.1yD@acsu.buffalo.edu> goetz@acsu.buffalo.edu (Phil Goetz) writes:
>On the contrary - the bot languages are too powerful for genetic algorithms.
>
>To use genetic improvement of algorithms, you have to be able to change the
>algorithm randomly.  A bot language is rather complex, and changing (say)
>a GOTO to a LET A = B would probably not improve performance.  You would have
>to sort through a lot of mutants before you found a viable one.
>
>The key to using genetic algorithms is to formalize the behavior/construction
>you wish to improve in a way such that a random change still produces something
>plausible.  I don't think that computer-controlled players can be generated
>using genetic algorithms.

Ok, maybe I'm off in left field, but in Steven Levy's _Artificial Life_
book, he talks about an experiment where programs fought to reproduce
themselves. I guess that I don't remember GAs mentioned there. However,
I started playing around with an old game that I had written that used
programmed players and I used GAs on their code. Now, at first I said
that since I was dealing with machine language and not genes, I would
get only garbage out. And indeed, I seeded the initial population with
some robots that I had written. Within 5 generations, my population was
uniform, and nearly identical to one of my seed robots.

However, I tried again (on a 486, this time :) and I have got some
interesting behaviors quite randomly. I think that I am going to have to
add a little bit more randomness into the brew, because things are
beginning to look too uniform again.

Ok, so to reply to the claim above, computer-controlled players CAN be
generated using genetic algorithms. I've been through 22 generations so
far, and haven't seen anything terrific (I guess that I'm still matching
the computer-generated 'bots to my own, and mine are better... now) but
these things are forming pretty much on their own.

>
>I am aware that a network can theoretically do any computable task.  However,
>if your function is not modularized well, the network will function
>inefficiently, to the point that it will not work.
>

Neural Nets cannot, to my understanding, compute multiplication of two
numbers. Perhaps you are not talking about NNs, or perhaps the
multiplication needs to be done differently than I had approached it. It
seems to me that NNs work fine when the outputs are a linear combination
of the inputs.


Just the $/50 of someone who's spent some spare time toying with this
stuff.

-- 
Your Mom!       The Carburator!         |I'm not paid to HAVE opinions.
Your Shoes!     A gardenhose!           |
You got to got to got to got to ...     |This sig can be yours; send $15 to
SCUD MISSILE!!!	-"Scud Missile", Grunge |tsmaster@athena.mit.edu