[computer-go] Thoughts about UCT (long with computer Go content)

[Benjamin Teuber]

Magnus Persson wrote:
> What makes this interesting is that if UCT makes a program stronger than alpha
> beta search using equivalent evaluation, then it may not be so because it
> searches more efficiently but because it has a different "philosophy" of
> selecting the best move, that may be particular useful for Go.
>   
I'm pretty sure the advantage comes mostly from having a way to grow the 
search-tree in an asymmetric way, although I can't consider myself an 
expert in monte-carlo algorithms.

Of course, it's also an advantage being able to take into account the 
uncertainty of calculations when chosing where to play, but I believe 
the former thing is a bigger advantage.

Regards,
Ben