Record-breaking win in computer Go match vs professional at 4 stones

gogameguru.com

13 points·mquander··6 comments

Some context: This is the first time a computer has beaten a professional in a slow game on a full-size board with a 4-stone handicap or less, putting the computer's strength above that of all but the most dedicated amateur players. Few people expected this result going into the match. It's also worth noting that the hardware was just a small cluster of four PCs, not extravagant.

Is there a paper behind the design of Zen? I'm assuming it's some combination of heuristics and MCTS.

I don't know of a paper; it's sold commercially by the author and is closed-source. It's definitely fair to say that it is a combination of MCTS and heuristics. The best I can find at a glance is this post on the computer-go list, where the author remarks:

Zen is similar to partly MoGo, partly Crazy Stone. It has shape patterns that are generated by minorization-maximization algorithm like Crazy Stone. But they are directly combined with UCT - I don't use progressive widening. Probably the most original part of Zen is in the playout. I don't think MC simulations must be always fast, so it has a lot of hard-coded Go knowledge.

http://www.mail-archive.com/computer-go@computer-go.org/msg1...

Is 30 minutes main time really a slow game for Go?

On KGS 25 minutes is actually considered a medium length game. Obviously real games between pros can go on for days, but for whatever reason most games with bots seem to be played at or near blitz speeds, so this is slow by comparison.

Interesting conclusion! nice post

Reply on news.ycombinator.com