[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: computer-go: Neural networks



Hi,

-----Ursprüngliche Nachricht-----
Von: Julian Churchill <jjc97c@xxxxxxxxxxxxxxxxx>
An: computer-go@xxxxxxxxxxxxxxxxx <computer-go@xxxxxxxxxxxxxxxxx>
Datum: Mittwoch, 16. Mai 2001 14:40
Betreff: computer-go: Neural networks


> I have just started a project to incorporate
> neural network techniques into a standard alpha beta
> search algorithm. I am wondering if anyone else has
> tried to apply neural networks to the domain of Go

my program Augos uses a 3 layer (one hidden)
feed forward network to recognize linkages between
chains, and a 4 layer (two hidden) network
to estimate the territorial value of a point.
Both nets learn with backpropagation.

Input of these nets is a couple of 3*3 vectors,
representing the point to evaluate, and the
sourrounding eight points context.

Each of those 3*3 vectors come from different
arrays that represent an aspect of the board:
Fields of force, strategic value, distance
to the edge, liberties.

Greetings from Goeppingen,
Joachim