[Computer-go] mcts and tactics
sheppardco at aol.com
Tue Dec 19 18:26:05 PST 2017
>I wouldn't find it so surprising if eventually the 20 or 40 block networks develop a set of convolutional channels that traces possible ladders diagonally across the board.
Learning the deep tactics is more-or-less guaranteed because of the interaction between search and evaluation through the training process.
Let’s look at the base case for tactical expertise: the evaluation function has learned that capturing stones is good.
In that state, a very short ladder can be seen through by tactics (800 nodes of search). This results in a corrective force that adjusts the evaluation function.
Once the evaluation function knows that very short ladders are favorable, then the search can see through deeper ladders.
This continues as the ladders become longer and longer.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Computer-go