Re: [computer-go] TD(lambda), Neural Networks and evaluation functions

["Nicol N. Schraudolph" <compgo@xxxxxxxxxxxxxxx>]

On Tue, 16 Sep 2003, Imran Ghory wrote:

> 1) Create 81 neural networks (one associated with each intersection on the
> board). Let's represent them by N(x, input-board) with x=1...81.
> 2) Use temporal difference learning to teach the neural networks, with the
> rewards being +1/-1 depending on which side controls that intersection at
> the end of that game.
> 
> Has anyone experimented with this kind of approach before ?

That's exactly what we did 10 years ago, except that we reduced the model
complexity by having all but the bias weights shared between the networks.
See http://n.schraudolph.org/pubs/gochap.pdf.  Yes it does work to some
extent, but you need better features to really get anywhere, a la NeuroGo.

- Nici.

-- 
    Dr. Nicol N. Schraudolph                 http://n.schraudolph.org/
    Steinwiesstr. 32                         mobile:  +41-76-585-3877
    CH-8032 Zurich, Switzerland                 tel:      -1-251-3661

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://computer-go.org/mailman/listinfo/computer-go