Dear Ingo and Pasky, >* the heavy-hardware one (project sponsored by Japanese company >Dwango). The Zen-Dwango project was initiated in March 2016, >just after the Lee-Sedol-AlphaGo match. It is highly ambitious.
Just one thing to correct. The project was announced at March 1st, before the match. >By the way, it would be interesting to see which other bots >have similar problems (like Zen19X) near the end of the round-1 game. The basics of MCTS is ramdom sampling from a set of (good) moves following to a leaf node of search tree, until game finishes. On (not only but most importantly) L&D problems, "tenuki" causes a big problem, ie, the probability of alive (or death) becomes not 100% but rather 80% or even 50%. On typical cases such as nakade, Zen is trying to make the prob 100% but in general cases it's impossible. So, for this game, center black group is killed in the simulation at certain percentage in (I geuss all) MC bots (including even AlphaGo which also uses MC simulations in part). Best, Hideki Ingo Althofer: <trinity-90fa9b4e-20ba-495b-9cd6-75360a6ce89d-1472539670396@3capp-gmx-bs23>: >Hi Petr, > >Hideki gave already a compact explanation. >Let me tell you another few lines. > >> Silly question - why not? (Assuming this seems to be the strongest Zen >> so far.) > >Your question is not silly at all, and a full answer would >be quite complicated. > >Currently there are two developZen19Xment lines of Zen: >* the traditional one (which runs on normal hardware, and from >which for instance the new commercail Zenith-6 stems) > >and > >* the heavy-hardware one (project sponsored by Japanese company >Dwango). The Zen-Dwango project was initiated in March 2016, >just after the Lee-Sedol-AlphaGo match. It is highly ambitious. > >Zen19X was winner of the KGS bot tournament in July 2016, and >it was our original intention that this winner should play >against the German champion. > >************************************* > >By the way, it would be interesting to see which other bots >have similar problems (like Zen19X) near the end of the round-1 game. > >Ingo. > >PS. I hope I did tell things in the right way. In particular, >Hideki may feel free to correct me in any point he wants. >_______________________________________________ >Computer-go mailing list >Computer-go@computer-go.org >http://computer-go.org/mailman/listinfo/computer-go -- Hideki Kato <mailto:hideki_ka...@ybb.ne.jp> _______________________________________________ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
