[computer-go] Monte-Carlo for Tactical Search

[Łukasz Lew]

Tristan Cazenave has some spectacular results for using MC to evaluate
goals like
connection, L&D and eye making.

I think this is a very good direction.

Best Regards,
Łukasz LEw

On 8/31/06, Benjamin Teuber <benjamin.teuber at web.de> wrote:
> It seems to me like all programs with Monto Carlo use it for full-board
> search only, which doesn't scale so well from 7x7 to 9x9, not to mention
> 19x19.
> I'd be rather interested if it could be applied for tactical local
> search like connection, capture, life&death etc. Given enough time for
> simpler problems, the algorithm should give an exact proof of the
> result; if this isn't possible, if could still give an estimation of how
> probable a capture seems.
> The interesting thing is how the locality can be assured - maybe the
> initial probabilities could depend on the distance from the
> search-target, for example...
>
> Any thoughts?
>
> Regards,
> Benjamin
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
>