Re: computer-go: abstract info and neural nets

[Heikki Levanto <heikki@xxxxxx>]

On Fri, Jan 11, 2002 at 03:45:36PM -0800, Tyler Gaede wrote:

> the neural network doesn't have to look at a whole board as a pattern.
> smaller patterns (of a 5x5 block, or maybe a 9x9 block) probably come up a
> lot more often and would be much more recognizable. in this way, instead of
> focusing on whole-board tactics, neural networks could help the "close
> combat" sequences.

That has its own drawbacks. Some "close combat" does not fit into a 5x5 or
9x9 matrix, occasionally a situation in one corner depends on a stone in the
opposite corner (ladder break). 

Another, more serious, objection is that even if the program would play
locally correct moves, it could never have an overall "understanding" of the
situation, and would thus generally play in the wrong area of the board. It
may make a beautiful connection between two dead groups, or carefully secure
a fourth eye to a group that is already unconditionally alive, or extend a
group into enemy territory, even if the group is dead as it stands, and
adding those stones just costs it points.


Or have I missed something again?

- Heikki

-- 
Heikki Levanto  LSD - Levanto Software Development   <heikki@xxxxxxxxxxxxxxxxx>