Re: computer-go: Programs learning to play Go

[Matthew Corey Brown <bromoc@xxxxxxxxxxxxxxxxxx>]

On Mon, 20 Aug 2001, Dave Dyer wrote:

> 
>  My observation on NN based learning is that the few cases where
> they have worked well involve situations where choosing one of a
> few alternatives based on a few inputs is the mode.  The prototype
> success story is Backgammon.
> 
>  Go doesn't really fit this model.  Throwing 361 inputs and outputs
> at a naive network, and expecting it to learn anything, is like
> throwing a pail of sand into an oven and expecting silicon chips
> to emerge.
The one thing in my opinion that makes it different from backgamonn and
checkers (blondie24) model, is passing.. THe other games there are a set
condition that determines the end of the game. Not so in Go, where the
players them selves choose when the game ends. (Well there is the rare
condition when there are no legal moves left) The trick is to seprate the
pasiing desicion from the best move descion. And this is what i am
planning once i finish the rewrite of my NN computation program.


> 
> #1: if you must try a pure learning technique, start with a reduced
> domain; 3x3 or 4x4 is a good place to start.  Achieve percect play 
> on 4x4, then work your way up.
Problem.. gnugo is really dumb on anything less then a 9x9 board, it
passes on the first move half the time. 


> The reason Go is so intractable to simple learning paragigms 
> is that many levels of abstraction are necessary to understand
> a position; and there are strong interactions both within and among
> abstraction layers. 
Put other intracatble problems (like TSP) are manageable with social
learning techniques. I do expect to take many 1000's of generations to
be even slightly competent as the dimensionality of the net is so
high. But social learning speeds up the training of nueral nets. (I expect
the 1000's of generations with social learning)


Matthew Corey Brown                             bromoc@xxxxxxxxxxxxxxxxx 
                  Happiness is a dry place to live.