Overblog Suivre ce blog
Editer l'article Administration Créer mon blog
4 mai 2011 3 04 /05 /mai /2011 16:02

MoGoTW, a program based on recent progresses in artificial intelligence (not using alpha-beta but Monte-Carlo Tree Search, see here for other applications of this beautiful techniques), won the first game with handicap 7

against a top professional player.

 

The graph of the game is here:

 

http://www.lri.fr/~teytaud/H7.jpeg

and the file of the game can be found here.

MoGoTW is supported by Grid5000, and was here using Huygens cluster. It benefitted also from support by NUTN

It was played in 2009 at Taiwan's open ( http://epochtimes.com/b5/9/2/11/n2425091.htm ).

 

The opponent is a top professional player, i.e. he is professional, ranked 9P (the highest level),

and recently won a major tournament (the LG Cup 2007):

http://img.epochtimes.com/i6/902101213481538--ss.jpg

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Left: Arpad Rimmel (MoGoTW's operator).

Right: Chun-Hsun Chou (9Dan Pro player).

 

 

  bibtex entry:

 

  @inproceedings{winWithH7AgainstPro,
    HAL_ID = {inria-00386477},
    URL = {http://hal.inria.fr/inria-00386477/en/},
    title = { {A}dding expert knowledge and exploration in {M}onte-{C}arlo {T}ree {S}earch},
    author = {{C}haslot, {G}uillaume and {F}iter, {C}hristophe and {H}oock, {J}ean-{B}aptiste and {R}immel, {A}rpad and {T}eytaud, {O}livier},
    abstract = {{W}e present a new exploration term, more efficient than clas- sical {UCT}-like exploration terms and combining efficiently expert rules, patterns extracted from datasets, {A}ll-{M}oves-{A}s-{F}irst values and classi- cal online values. {A}s this improved bandit formula does not solve several important situations (semeais, nakade) in computer {G}o, we present three other important improvements which are central in the recent progress of our program {M}o{G}o: { {W}e show an expert-based improvement of {M}onte-{C}arlo simulations for nakade situations; we also emphasize some limitations of this modification. { {W}e show a technique which preserves diversity in the {M}onte-{C}arlo simulation, which greatly improves the results in 19x19. { {W}hereas the {UCB}-based exploration term is not efficient in {M}o{G}o, we show a new exploration term which is highly efficient in {M}o{G}o. {M}o{G}o recently won a game with handicap 7 against a 9{D}an {P}ro player, {Z}hou {J}un{X}un, winner of the {LG} {C}up 2007, and a game with handicap 6 against a 1{D}an pro player, {L}i-{C}hen {C}hien.},
    language = {{E}nglish},
    affiliation = {{M}aastricht {U}niversity - univ. {M}aastricht - {TAO} - {INRIA} {S}aclay - {I}le de {F}rance - {INRIA} - {CNRS} : {UMR}8623 - {U}niversit{\'e} {P}aris {S}ud - {P}aris {XI} - {TAO} - {INRIA} {F}uturs - {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}aris {XI} },
    booktitle = {{A}dvances in {C}omputer {G}ames },
    publisher = {{S}pringer },
    address = {{P}amplona {S}pain },
    audience = {international },
    year = {2009},
    URL = {http://hal.inria.fr/inria-00386477/PDF/peacg.pdf},
}

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Partager cet article

Repost 0
Published by teytaud - dans Recherche
commenter cet article

commentaires

Buy Articles 18/06/2011 00:24


I find this information very useful and it has considerably saved my time.thanks.