Re: [computer-go] Pattern matching - example play

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 02:37 4-12-2004 +0100, Arend Bayer wrote:
>
>On Fri, 3 Dec 2004, Vincent Diepeveen wrote:
>
>> >I would not completely write off higher-level planning and neural nets,
>> >and other fancy theories. Many of them have shown their values in
>> 
>> I have written them off after i learned a lot about them and experimented
>> with them myself. Additional it won't be a surprise to you that worlds
>> biggest ANN experts are chatting to the programmers so also to me and they
>> can exactly define the problem for you.
>> 
>> The problem is that game playing programs each time play an unique game. So
>> they should train in order to recognize something they didn't train at!
>> 
>> That's a contradiction and therefore it is not possible.
>
>> We can have lengthy discussions, but the majority of ANN top researchers
>> agree with me here that for game playing ANN is completely useless.
>
>So you would claim that neural nets are worthless for game like, say,
>backgammon?

Backgammon is a statistical game where statistics drive the game.

It is not a game tree search like go or chess are.

It is a statistical tree you work out in backgammon. 

You are dependant upon the dices. Playing very stubbornly at statistics has
its merits then and usually wins the game. 

I do not see why a neural net wouldn't be able to learn statistics, as 2
dices have just 36 possibilities each time with a chance attached.

So it can learn about those 36 possibilities and each time those statistics
are true again, because it is the same 36 possibilities.

There is no situation change. No new pattern. The statistics are the same
for each dice combination again.

Add to that, that the branching factor is so huge and that people love to
play for money against each other and not against a computer.

Certain persons who earned their fortune with oil, to say it in a polite
manner, are known for playing in total for millions such games.

So it is the dices that makes it interesting for them to gamble in the game.

The best backgammon players happen to be chessplayers by the way (and no
doubt many go players in Asia too).

Which in itself is amazing, i would have personally expected to see a lot
of bridge players there too as card players are real good in working with
statistics. 

Perhaps the signalling aspect of what your partner has in bridge is far
more important than we are all suspecting, which makes them worthless in
backgammon.

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