After this unbelievable streak of 60 won games (even though we still have to see how it holds up with longer time control) it’s not completely unthinkable anymore to play top pros with a handicap. Sadly because the Komi is fixed for the value network it seems the next bigger handicap is 2 stones with Komi for white which seems a big jump. Also having 2 stone games is not so interesting since it would reveal less insights for even game opening Theory. So my question is: is it possible to have reverse Komi games by feeding the value network with reverse colors? Or wouldn’t that work because the value network is so fine calibrated that it would throw it off if there is one more white stone than usual? I guess the policy network and MTSC should have no problems with a changed Komi?
Maybe Aja or some other expert for Value networks can answer that? Thanks David O. > On Jan 4, 2017, at 8:02 AM, Jim O'Flaherty <jim.oflaherty...@gmail.com> wrote: > > Tysvm for posting that! > > I had predicted it was AlphaGo from the beginning. If there is a competitor > emerging, I think we would have seen some sort of publicity around it, if not > just to provoke a response with the AlphaGo team. > > > On Wed, Jan 4, 2017 at 9:11 AM, Janzert <janz...@janzert.com > <mailto:janz...@janzert.com>> wrote: > On 1/2/2017 7:05 AM, "Ingo Althöfer" wrote: > Hello Paweł, > > There have been another 8 games on Foxwq server: > ... > Totally, 38-0. It looks like a kind, indirect (yet powerful), message > from DeepMind to Chinese Go Association: "Please, let us try a real > challenge, like 3-handicap games, it does not really make much sense > to play even anymore". > > So, do you want to say that "Master" might be AlphaGo? > From the disucssion I thought that "Master" was a chinese bot. > > If Aja is reading: can you enlighten us? > > Cheers, Ingo. > > Looks like we have an official answer in the affirmative > https://twitter.com/demishassabis/status/816660463282954240 > <https://twitter.com/demishassabis/status/816660463282954240> > > _______________________________________________ > Computer-go mailing list > Computer-go@computer-go.org <mailto:Computer-go@computer-go.org> > http://computer-go.org/mailman/listinfo/computer-go > <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
