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. If it had enough examples of ladders of different lengths, including selfplay games where game-critical ladders "failed to be understood" by one side or the other and possibly even got played out, it seems like the neural net would have a significant incentive to learn them, step by step.
On Tue, Dec 19, 2017 at 7:57 PM, Andy <andy.olsen...@gmail.com> wrote: > How do you interpret this quote from the AGZ paper? > "Surprisingly, shicho (“ladder” capture sequences that may span the whole > board) – one of the first elements of Go knowledge learned by humans – were > only understood by AlphaGo Zero much later in training." > > To me "understood" means the neural network itself can read at least some > simple whole board ladders, ladder breakers, and ladder makers. I would > find it a large oversell if they just mean the MCTS search reads the ladder > across the whole board. > > > > 2017-12-19 18:16 GMT-06:00 Stephan K <stephan.ku...@gmail.com>: > >> 2017-12-20 0:26 UTC+01:00, Dan <dsha...@gmail.com>: >> > Hello all, >> > >> > It is known that MCTS's week point is tactics. How is AlphaZero able to >> > resolve Go tactics such as ladders efficiently? If I recall correctly >> many >> > people were asking the same question during the Lee Sedo match -- and it >> > seemed it didn't have any problem with ladders and such. >> >> Note that the input to the neural networks in the version that played >> against Lee Sedol had a lot of handcrafted features, including >> information about ladders. See "extended data table 2", page 11 of the >> Nature article. You can imagine that as watching the go board through >> goggles that put a flag on each intersection that would result in a >> successful ladder capture, and another flag on each intersection that >> would result in a successful ladder escape. >> >> (It also means that you only need to read one move ahead to see >> whether a move is a successful ladder breaker or not.) >> >> Of course, your question still stands for the Zero versions. >> >> Here is the table : >> >> Feature # of planes Description >> >> Stone colour 3 Player stone / >> opponent stone / empty >> Ones 1 A constant plane >> filled with 1 >> Turns since 8 How many turns >> since a move was played >> Liberties 8 Number of >> liberties (empty adjacent points) >> Capture size 8 How many opponent >> stones would be captured >> Self-atari size 8 How many of own >> stones would be captured >> Liberties after move 8 Number of >> liberties after this move is played >> Ladder capture 1 Whether a move at this >> point is a successful ladder capture >> Ladder escape 1 Whether a move at >> this point is a successful ladder escape >> Sensibleness 1 Whether a move is >> legal and does not fill its own eyes >> Zeros 1 A constant plane >> filled with 0 >> >> Player color 1 Whether current >> player is black >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org >> http://computer-go.org/mailman/listinfo/computer-go > > > > _______________________________________________ > Computer-go mailing list > Computer-go@computer-go.org > http://computer-go.org/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
