Mardi 20 juillet 2010 2 20 /07 /Juil /2010 13:49

 

The 20th of July, 2010, at WCCI 2010 in Barcelona, some games were played by boths against humans.

This page is centered on the games played by MoGo and MoGoTW, but I give also a fast overview of other game.

A main novelty is the presence of 13x13 games. MoGo and MoGoTW are supported by the Grid5000 project.


There are not so many games against strong humans and computers in 13x13; I guess the game below is the first ever win of a computer against a 6D human in 13x13 Go with handicap 2:

 

http://www.lri.fr/~teytaud/blo1313.jpg

 

 



This game was won by MoGo playing on 15 octo-cores. As usual, the bot gave away some points at the end, and won just by 0.5 whereas winning by more points was easy.The SGF file can be found here:

http://www.lri.fr/~teytaud/ARCHIVE_BARCELONA/firstEverWinAgainst6D_with_Handicap2_in_13x13.sgf

 

MoGo won the next game with H2 against another 6D player, Shang-Rong Ts

ai, also in 13x13:

 

13x13H2b.jpg

 

At this point, MoGo (black) used a threat in B2; white can only win the ko (F19) by playing in the upper part (G11), otherwise black wins the game by killing the big white group H13; so white does not reply to the threat, MoGo captures the lower left corner and black (MoGo) wins. This shows that bots can sometimes play well ko fights.The complete game

can be found here:

 

http://www.lri.fr/~teytaud/ARCHIVE_BARCELONA/win_By_MoGo_against_Shang-Rong_Tsai_6D_13x13_H2___wellDoneKoFight.sgf

 

 

 

Incidentally, MoGo won a 9x9 game against a 9P as white:

 

http://www.lri.fr/~teytaud/b99a.jpg

 

 



The pro said that move 28 was a very good move, necessary for winning this game. This makes MoGoTW the first ever bot which won against a pro both a black and as white. The game can be found here:

http://www.lri.fr/~teytaud/ARCHIVE_BARCELONA/gameWonIn9x9AsWhiteAgainst9P.sgf


On the other hand, MoGo lost as black, in spite of an opening very close to a past win
against a 9P player:

 

http://www.lri.fr/~teytaud/b99b.jpg

 

 




Other bots:

  • Many Faces of Go also won a game against a 6D with H2, a few hours later; it was against Shi-Jim Yen. Many Faces then lost against Shang-Rong Tsai in the same configuration (6D, 13x13, H2).
  • Fuego won one of two games against a 4P player in 9x9
  • Fuego lost twice with H2 against a 6D, showing that it's not so easy to win against strong humans with H2 in 13x13.
  • Zen won thee games out of four in 9x9 against a 6D, confirming that computers are now at a professional level in 9x9.
  • Zen and ManyFaces lost in 19x19 with handicap 7 and 6 respectively against a 9P and a 4P respectively in the morning, showing that we need many improvements for competing with humans in 19x19; but in the afternoon, if ManyFaces lost again, with H7 against the 9P, Zen could win with H6 against the 4P player in 19x19. This was a win by time, but the game was in favor of black (Zen). Congratulations to Zen and its developpers! The game can be found in the KGS archive, 20th of July 2010, user  Zen19 around 3pm GMT (the last game that day).

 

Paper on technical tools for this: http://hal.archives-ouvertes.fr/inria-00544622/en/

 

bibtex entry:

  @article{RIMMEL:2010:INRIA-00544622:1,
        HAL_ID = {inria-00544622},
            URL = {http://hal.archives-ouvertes.fr/inria-00544622/en/},
            title = { {C}urrent {F}rontiers in {C}omputer {G}o},
                author = {{R}immel, {A}rpad and {T}eytaud, {O}livier and {L}ee, {C}hang-{S}h
ing and {Y}en, {S}hi-{J}im and {W}ang, {M}ei-{H}ui and {T}sai, {S}hang-{R}ong},
                abstract = {{T}his paper presents the recent technical advances in {M}on
te-{C}arlo {T}ree {S}earch for the {G}ame of {G}o, shows the many similarities and the rare differen
ces between the current best programs, and reports the results of the computer-{G}o event organized
at {FUZZ}-{IEEE} 2009, in which four main {G}o programs played against top level humans. {W}e see th
at in 9x9, computers are very close to the best human level, and can be improved easily for the open
ing book; whereas in 19x19, handicap 7 is not enough for the computers to win against top level prof
essional players, due to some clearly understood (but not solved) weaknesses of the current algorith
ms. {A}pplications far from the game of {G}o are also cited. {I}mportantly, the first ever win of a c
omputer against a 9th {D}an professional player in 9x9 {G}o occurred in this event.},
                    language = {{A}nglais},
                    affiliation = {{D}epartment of {C}omputing {S}cience - {D}epartm
ent of {C}omputing {S}cience, {U}niversity of {A}lberta - {TAO} - {INRIA} {S}aclay - {I}le de {F}ran
ce - {INRIA} - {CNRS} : {UMR}8623 - {U}niversit{\'e} {P}aris {S}ud - {P}aris {XI} - {L}aboratoire de
 {R}echerche en {I}nformatique - {LRI} - {CNRS} : {UMR}8623 - {U}niversit{\'e} {P}aris {S}ud - {P}ar
is {XI} - {D}epartment of {C}omputer {S}cience and {I}nformation {E}ngineering - {NUTN} - {N}ational
 {D}ong {H}wa {U}niversity - {NDHU} - {NDHU} - {D}epartment of {I}nformation {M}anagement - {CJCU} }
,
                        pages = {in press },
                        journal = {{IEEE} {T}ransactions on {C}omputational {I}n
telligence and {A}rtificial {I}ntelligence in {G}ames },
                            audience = {internationale },
                            year = {2010},
                                URL = {http://hal.archives-ouvertes.fr/inria
-00544622/PDF/ct.pdf},
}

Par teytaud - Publié dans : Recherche - Communauté : Science & Avenir
Ecrire un commentaire - Voir les 0 commentaires
Retour à l'accueil
Créer un blog gratuit sur over-blog.com - Contact - C.G.U. - Rémunération en droits d'auteur - Signaler un abus - Articles les plus commentés