[MUD-Dev] Neural Networks as the AI system for a MUD?

[Kræn Munck <fluid@bollektivet.dk>]

From: "Marc Hernandez" <marc at ias.jb.com>
> Actually, NNs have been used successfully to beat other AI
> techniques in at least one realm, backgammon.  TD-Gammon uses the NN, I
> believe, to select the next move, and Temporal Difference learning to
> train it.  TD is a nifty AI technique in and of itself.

You're right about TD being a very powerful technique. I've had some rather
nice results applying it to Go. But it is very expensive on the computation
side, as it (often) requires a lot of neurons to represent the output space.
In a situation where you have a couple of hundred entities wandering around,
I don't think it's a realistic alternative.

/Kræn Munck




_______________________________________________
MUD-Dev maillist  -  MUD-Dev at kanga.nu
http://www.kanga.nu/lists/listinfo/mud-dev