• 微博
  • 微信微信二维码

广东省人民政府首页  >  要闻动态  >  广东要闻

德州扑克人工智能之父:阿尔法狗赢不赢都缺实际用处

来源: 南方日报网络版     时间: 2019年05月22日 11:58:56
【字体:

开封开会议费发票██【电\微 15813★310971 李先生】【Q; 25880★8054】 ██▓信誉100%,专业100%,选择我们长期合作共赢!公司经百度推广!德州扑克人工智能之父:阿尔法狗赢不赢都缺实际用处

朱德弥留之际嘱咐李先念 一句话改变国人命运-搜狐新闻 毛泽东临终之际 紧握叶剑英手放心不下一件事-搜狐新闻 霍英东弥留之际 胡锦涛曾前往探望(组图)-搜狐新闻 患癌女星徐婷最后时光:弥留之际默念阿弥陀佛-搜狐新闻 罗青长之子忆沈安娜:弥留之际仍担心暴露-搜狐军事频道 赵乐际任中组部部长 不再兼任陕西省委书记(图)-搜狐新闻 习近平:既开动脑筋又不脱离实际好高骛远-搜狐新闻 习近平:脱贫不要脱离实际随意提前 -搜狐新闻 上海交通大学建校120周年 江泽民致贺信-搜狐新闻 全国实际吸毒人数估计超1400万 100人中就有1人_手机搜狐网 105岁老红军寇庆延遗体告别 习近平江泽民胡锦涛哀悼_同志 雷曼兄弟申请破产-搜狐新闻 写在习近平总书记视察安徽一周年之际-搜狐新闻 美曝中国洲际导弹库 火力覆盖大半个地球(图)-搜狐军事频道 陈毅之子陈小鲁回应:我不是安邦实际控制人-搜狐新闻

(  人)(机)(对)(战)(无)(疑)(是)(4月)(10日)(的)(新)(闻)(焦)(点)(之)(一)(。一)(边)(在)(海)(南)(澄)(迈)(,德)(州)(扑)(克)(人)(工)(智)(能)(“冷)(扑)(大)(师)(”(Libratus)经)(过)(5日)(激)(战)(,在)(33000手)(一)(对)(一)(无)(限)(注)(德)(扑)(比)(赛)(中)(完)(胜)(中)(国)(“龙)(之)(队)(”的)(6名)(高)(手)(,斩)(获)(200万)(人)(民)(币)(奖)(金)(;另)(一)(边)(在)(北)(京)(的)(中)(国)(棋)(院)(,谷)(歌)(宣)(布)(被)(中)(国)(网)(民)(称)(为)(“阿)(尔)(法)(狗)(”的)(围)(棋)(人)(工)(智)(能)(AlphaGo将)(在)(5月)(下)(旬)(的)(乌)(镇)(对)(阵)(柯)(洁)(领)(衔)(的)(中)(国)(顶)(尖)(棋)(手)(,上)(演)(终)(极)(一)(战)(。  一)(南)(一)(北)(,一)(牌)(一)(棋)(,两)(则)(人)(机)(对)(战)(重)(磅)(消)(息)(在)(同)(一)(天)(出)(炉)(,虽)(然)(未)(必)(是)(有)(意)(针)(锋)(相)(对)(,但)(也)(足)(够)(令)(无)(心)(的)(观)(众)(相)(互)(比)(较)(。有)(些)(读)(者)(认)(为)(,连)(围)(棋)(这)(样)(高)(深)(的)(智)(力)(运)(动)(都)(早)(已)(被)(阿)(尔)(法)(狗)(拿)(下)(,扑)(克)(这)(样)(门)(槛)(低)(、上)(手)(简)(单)(的)(大)(众)(娱)(乐)(又)(何)(足)(挂)(齿)(呢)(?但)(反)(方)(则)(指)(出)(,扑)(克)(不)(是)(单)(纯)(靠)(计)(算)(的)(游)(戏)(,牵)(扯)(到)(很)(多)(心)(理)(上)(的)(博)(弈)(,有)(时)(甚)(至)(需)(要)(“飙)(演)(技)(”,AI能)(掌)(握)(这)(类)(游)(戏)(有)(些)(匪)(夷)(所)(思)(。  那)(么)(,德)(扑)(和)(围)(棋)(这)(两)(种)(智)(力)(游)(戏)(,对)(AI来)(说)(到)(底)(有)(什)(么)(区)(别)(?哪)(个)(难)(度)(更)(大)(一)(些)(?针)(对)(这)(些)(问)(题)(,澎)(湃)(新)(闻)((www.thepaper.cn)就)(冷)(扑)(大)(师)(和)(阿)(尔)(法)(狗)(的)(相)(关)(资)(料)(进)(行)(了)(梳)(理)(与)(比)(较)(。托)(马)(斯)(·桑)(德)(霍)(姆)((TuomasSandholm)资)(料)(图)(  不)(过)(,冷)(扑)(大)(师)(的)(开)(发)(者)(、美)(国)(卡)(内)(基)(梅)(隆)(大)(学)((CMU)计)(算)(机)(系)(教)(授)(托)(马)(斯)(·桑)(德)(霍)(姆)((TuomasSandholm)在)(听)(到)(阿)(尔)(法)(狗)(将)(来)(华)(挑)(战)(的)(消)(息)(后)(,评)(价)(非)(常)(的)(简)(单)(:问)(题)(的)(关)(键)(并)(不)(在)(于)(中)(国)(顶)(尖)(棋)(手)(能)(不)(能)(赢)(,而)(在)(于)(生)(活)(中)(几)(乎)(不)(存)(在)(像)(围)(棋)(这)(样)(的)(“完)(美)(信)(息)(游)(戏)(”,因)(而)(阿)(尔)(法)(狗)(也)(缺)(乏)(实)(用)(场)(景)(。  围)(棋)(和)(德)(扑)(对)(于)(AI来)(说)(,性)(质)(是)(不)(是)(一)(样)(的)(?  自)(1988年)(CMU开)(发)(的)(“奥)(赛)(罗)(”程)(序)(击)(败)(世)(界)(黑)(白)(棋)(冠)(军)(以)(来)(,人)(类)(就)(在)(棋)(类)(游)(戏)(上)(节)(节)(败)(退)(。1997年)(,IBM的)(计)(算)(机)(“深)(蓝)(”击)(败)(了)(国)(际)(象)(棋)(棋)(坛)(神)(话)(卡)(斯)(帕)(罗)(夫)(,使)(人)(机)(对)(战)(在)(当)(时)(达)(到)(了)(舆)(论)(高)(峰)(,其)(影)(响)(力)(绝)(不)(亚)(于)(去)(年)(阿)(尔)(法)(狗)(击)(败)(韩)(国)(棋)(手)(李)(世)(石)(。  计)(算)(机)(程)(序)(攻)(克)(这)(些)(棋)(类)(游)(戏)(的)(方)(法)(在)(本)(质)(上)(是)(一)(样)(的)(:搜)(索)(。下)(每)(一)(步)(棋)(都)(是)(一)(个)(决)(策)(点)(,在)(决)(策)(点)(上)(分)(化)(出)(不)(同)(的)(可)(能)(性)(,而)(整)(盘)(棋)(也)(就)(形)(成)(了)(一)(棵)(不)(停)(分)(叉)(的)(决)(策)(树)(。而)(计)(算)(机)(所)(做)(的)(,就)(是)(在)(每)(一)(个)(决)(策)(点)(上)(搜)(索)(胜)(率)(最)(大)(的)(那)(条)(路)(径)(。  因)(而)(难)(度)(主)(要)(取)(决)(于)(决)(策)(点)(的)(数)(量)(。决)(策)(点)(越)(多)(,搜)(索)(难)(度)(自)(然)(越)(大)(。黑)(白)(棋)(的)(搜)(索)(量)(只)(有)(10的)(十)(几)(次)(方)(,国)(际)(象)(棋)(是)(10的)(四)(十)(几)(次)(方)(,而)(围)(棋)(的)(搜)(索)(量)(达)(到)(了)(惊)(人)(的)(10的)(170次)(方)(。  不)(过)(,也)(正)(因)(为)(围)(棋)(的)(搜)(索)(量)(太)(大)(,完)(全)(精)(确)(的)(结)(果)(是)(不)(可)(能)(得)(到)(的)(。因)(而)(,阿)(尔)(法)(狗)(在)(蒙)(特)(卡)(洛)(树)(搜)(索)(算)(法)((MonteCarloTreeSearch)的)(基)(础)(上)(,结)(合)(走)(棋)(网)(络)((PolicyNetwork)、快)(速)(走)(子)((Fastrollout)和)(价)(值)(网)(络)((ValueNetwork)进)(行)(预)(测)(和)(评)(估)(,得)(到)(尽)(可)(能)(接)(近)(最)(优)(的)(解)(。虽)(然)(阿)(尔)(法)(狗)(这)(个)(过)(程)(中)(做)(出)(了)(巨)(大)(突)(破)(,但)(它)(本)(质)(上)(还)(是)(在)(搜)(索)(。  德)(州)(扑)(克)(与)(棋)(类)(完)(全)(不)(同)(。围)(棋)(是)(一)(种)(“完)(美)(信)(息)(游)(戏)(”,场)(面)(上)(所)(有)(的)(情)(况)(都)(黑)(白)(分)(明)(地)(摆)(在)(棋)(面)(上)(,双)(方)(掌)(握)(对)(等)(的)(信)(息)(。然)(而)(,德)(扑)(中)(隐)(藏)(了)(很)(多)(信)(息)(,是)(一)(种)(“非)(完)(美)(信)(息)(游)(戏)(”。玩)(家)(不)(知)(道)(对)(手)(手)(中)(是)(什)(么)(牌)(,不)(知)(道)(五)(张)(公)(共)(牌)(会)(开)(出)(怎)(样)(的)(结)(果)(,也)(不)(知)(道)(对)(手)(猜)(测)(自)(己)(握)(有)(怎)(样)(的)(手)(牌)(。  因)(此)(,虽)(然)(一)(盘)(德)(扑)(的)(决)(策)(点)(数)(量)(是)(10的)(160次)(方)(,要)(少)(于)(围)(棋)(,但)(光)(靠)(搜)(索)(是)(打)(不)(了)(德)(扑)(的)(。德)(扑)(由)(此)(跳)(出)(了)(在)(搜)(索)(量)(上)(的)(较)(量)(,对)(AI提)(出)(了)(另)(一)(个)(方)(向)(上)(的)(考)(验)(:应)(对)(隐)(藏)(的)(信)(息)(。  阿)(尔)(法)(狗)(和)(冷)(扑)(大)(师)(的)(原)(理)(有)(何)(不)(同)(?  谷)(歌)(曾)(在)(《自)(然)(》杂)(志)(上)(发)(表)(文)(章)(,公)(布)(阿)(尔)(法)(狗)(的)(几)(个)(基)(本)(原)(理)(,分)(别)(为)(:走)(棋)(网)(络)((PolicyNetwork),给)(定)(当)(前)(局)(面)(,预)(测)(和)(采)(样)(下)(一)(步)(的)(走)(棋)(;快)(速)(走)(子)((Fastrollout),在)(适)(当)(牺)(牲)(走)(棋)(质)(量)(的)(条)(件)(下)(提)(高)(速)(度)(;价)(值)(网)(络)((ValueNetwork),给)(定)(当)(前)(局)(面)(,估)(计)(双)(方)(胜)(率)(;蒙)(特)(卡)(罗)(树)(搜)(索)((MonteCarloTreeSearch,MCTS),把)(以)(上)(三)(个)(部)(分)(串)(联)(成)(一)(个)(完)(整)(的)(系)(统)(。  冷)(扑)(大)(师)(的)(算)(法)(则)(主)(要)(基)(于)(以)(下)(三)(个)(模)(块)(产)(生)(:  首)(先)(,纳)(什)(均)(衡)(是)(德)(扑)(算)(法)(的)(核)(心)(,即)(AI的)(目)(标)(是)(找)(到)(一)(个)(无)(论)(对)(方)(怎)(么)(做)(,自)(己)(都)(不)(会)(产)(生)(损)(失)(的)(策)(略)(。根)(据)(博)(弈)(论)(,像)(一)(对)(一)(扑)(克)(这)(种)(零)(和)(游)(戏)(永)(远)(存)(在)(这)(样)(的)(最)(优)(解)(,就)(像)(在)(经)(典)(的)(博)(弈)(论)(模)(型)(“囚)(徒)(困)(境)(”中)(,招)(认)(罪)(行)(就)(是)(一)(个)(无)(论)(对)(方)(招)(认)(不)(招)(认)(都)(最)(优)(的)(策)(略)(。因)(此)(,冷)(扑)(大)(师)(的)(开)(发)(团)(队)(提)(前)(在)(庞)(大)(的)(决)(策)(树)(上)(利)(用)(虚)(拟)(遗)(憾)(最)(小)(化)(算)(法)((CounterfactualRegretMinimization,CFR)推)(算)(出)(了)(均)(衡)(,即)(通)(过)(多)(次)(迭)(代)(计)(算)(博)(弈)(树)(中)(每)(个)(信)(息)(集)(的)(动)(作)(遗)(憾)(值)(和)(平)(均)(策)(略)(值)(,预)(测)(下)(一)(时)(刻)(的)(决)(策)(动)(作)(,使)(其)(是)(当)(前)(最)(小)(遗)(憾)(动)(作)(。  此)(外)(,冷)(扑)(大)(师)(还)(有)(残)(局)(解)(算)(器)((end-gamesolver)和)(自)(我)(强)(化)(学)(习)(这)(两)(个)(模)(块)(来)(辅)(助)(第)(一)(个)(模)(块)(。残)(局)(解)(算)(器)(会)(在)(残)(局)(时)(实)(时)(评)(估)(场)(上)(的)(情)(况)(,以)(判)(断)(第)(一)(模)(块)(中)(算)(出)(的)(纳)(什)(均)(衡)(是)(否)(符)(合)(实)(时)(情)(况)(。而)(自)(我)(强)(化)(学)(习)(会)(反)(思)(AI在)(比)(赛)(中)(的)(表)(现)(,找)(出)(曾)(被)(人)(类)(利)(用)(过)(的)(“套)(路)(”,清)(除)(这)(些)(可)(循)(的)(痕)(迹)(。  总)(体)(来)(说)(,阿)(尔)(法)(狗)(和)(冷)(扑)(大)(师)(的)(算)(法)(有)(三)(大)(明)(显)(不)(同)(。  第)(一)(,阿)(尔)(法)(狗)(的)(训)(练)(过)(程)(中)(用)(到)(了)(大)(量)(数)(据)(的)(深)(度)(学)(习)(,即)(通)(过)(大)(量)(给)(定)(的)(输)(入)(和)(输)(出)(形)(成)(稳)(定)(的)(处)(理)(。但)(计)(算)(机)(的)(处)(理)(过)(程)(本)(身)(是)(一)(个)(黑)(匣)(子)(。而)(冷)(扑)(大)(师)(没)(有)(用)(到)(时)(髦)(的)(深)(度)(学)(习)(,而)(是)(通)(过)(传)(统)(的)(线)(性)(规)(划)(提)(前)(算)(出)(纳)(什)(均)(衡)(,是)(一)(个)(“老)(式)(但)(好)(用)(的)(人)(工)(智)(能)(”(GoodOld-FashionedArtificialIntelligence)。  第)(二)(,阿)(尔)(法)(狗)(在)(深)(度)(学)(习)(过)(程)(中)(参)(考)(了)(数)(千)(万)(张)(人)(类)(棋)(谱)(,可)(以)(说)(是)(靠)(模)(仿)(人)(类)(高)(手)(起)(步)(的)(。虽)(然)(阿)(尔)(法)(狗)(也)(运)(用)(到)(了)(一)(些)(自)(我)(强)(化)(学)(习)(,但)(比)(重)(并)(不)(大)(。而)(冷)(扑)(大)(师)(完)(全)(是)(通)(过)(自)(我)(强)(化)(学)(习)(进)(行)(训)(练)(的)(。开)(发)(者)(从)(未)(教)(给)(它)(人)(类)(打)(牌)(的)(方)(法)(,只)(是)(向)(它)(描)(述)(了)(德)(扑)(的)(规)(则)(,由)(它)(“左)(右)(互)(搏)(”,摸)(索)(出)(德)(扑)(应)(该)(怎)(么)(玩)(。因)(而)(,冷)(扑)(大)(师)(的)(打)(法)(完)(全)(脱)(离)(了)(人)(类)(经)(验)(。值)(得)(注)(意)(的)(是)(,即)(将)(与)(柯)(洁)(在)(乌)(镇)(对)(战)(的)(阿)(尔)(法)(狗)(2.0将)(摈)(弃)(人)(类)(棋)(谱)(,完)(全)(采)(用)(一)(套)(自)(我)(摸)(索)(出)(来)(的)(下)(法)(。  第)(三)(,虽)(然)(阿)(尔)(法)(狗)(和)(冷)(扑)(大)(师)(都)(特)(别)(喜)(欢)(残)(局)(阶)(段)(下)(杀)(招)(,但)(它)(们)(的)(难)(度)(不)(一)(样)(。围)(棋)(下)(到)(残)(局)(,可)(走)(的)(招)(数)(越)(来)(越)(少)(,计)(算)(量)(也)(越)(来)(越)(简)(单)(;而)(德)(扑)(开)(到)(转)(牌)(和)(河)(牌)(时)(,可)(能)(性)(更)(多)(,局)(面)(变)(得)(更)(为)(复)(杂)(。因)(此)(,冷)(扑)(大)(师)(经)(常)(在)(转)(牌)(阶)(段)(做)(出)(长)(时)(间)(的)(停)(顿)(。  阿)(尔)(法)(狗)(和)(冷)(扑)(大)(师)(哪)(个)(更)(有)(用)(?  在)(冷)(扑)(大)(师)(的)(开)(发)(者)(、美)(国)(卡)(内)(基)(梅)(隆)(大)(学)((CMU)计)(算)(机)(系)(教)(授)(托)(马)(斯)(·桑)(德)(霍)(姆)((TuomasSandholm)和)(其)(博)(士)(生)(诺)(姆)(·布)(朗)((NoamBrown)看)(来)(,这)(个)(问)(题)(的)(答)(案)(十)(分)(明)(显)(。在)(现)(实)(生)(活)(中)(,几)(乎)(不)(存)(在)(像)(围)(棋)(这)(样)(给)(出)(所)(有)(信)(息)(的)(完)(美)(情)(境)(,因)(而)(像)(阿)(尔)(法)(狗)(这)(样)(的)(算)(法)(也)(很)(难)(直)(接)(派)(上)(用)(场)(。  在)(现)(实)(生)(活)(中)(,我)(们)(遇)(到)(的)(事)(情)(会)(更)(像)(玩)(德)(扑)(:商)(业)(谈)(判)(时)(对)(方)(未)(知)(的)(底)(牌)(、房)(屋)(拍)(卖)(时)(竞)(争)(对)(手)(难)(测)(的)(举)(动)(、股)(票)(交)(易)(中)(一)(些)(隐)(藏)(的)(内)(部)(消)(息)(。桑)(德)(霍)(姆)(认)(为)(,德)(扑)(AI打)(开)(了)(靠)(人)(工)(智)(能)(解)(决)(随)(机)(事)(件)(和)(隐)(藏)(信)(息)(的)(大)(门)(。这)(样)(的)(AI,才)(有)(望)(离)(开)(虚)(拟)(的)(棋)(牌)(世)(界)(,成)(为)(人)(类)(在)(现)(实)(生)(活)(中)(谈)(判)(、博)(弈)(和)(投)(资)(的)(好)(帮)(手)(。  桑)(德)(霍)(姆)(自)(己)(就)(成)(立)(了)(一)(家)(战)(略)(计)(算)(公)(司)(,希)(望)(利)(用)(人)(工)(智)(能)(解)(决)(金)(融)(上)(的)(一)(些)(战)(略)(计)(算)(问)(题)(。首)(先)(,金)(融)(交)(易)(拼)(的)(是)(速)(度)(,AI在)(量)(化)(交)(易)(上)(具)(有)(巨)(大)(优)(势)(。其)(次)(,像)(股)(票)(交)(易)(中)(经)(常)(会)(存)(在)(一)(些)(隐)(藏)(的)(信)(息)(,AI能)(在)(交)(易)(者)(较)(少)(的)(情)(况)(下)(,帮)(助)(人)(类)(在)(隐)(藏)(的)(信)(息)(面)(前)(做)(出)(决)(策)(。  阿)(尔)(法)(狗)(和)(冷)(扑)(大)(师)(哪)(个)(更)(厉)(害)(?  诺)(姆)(·布)(朗)(说)(道)(,每)(一)(种)(游)(戏)(从)(计)(算)(机)(科)(学)(的)(角)(度)(来)(说)(,都)(存)(在)(一)(些)(核)(心)(问)(题)(,而)(掌)(握)(这)(个)(游)(戏)(的)(AI相)(当)(于)(解)(决)(了)(这)(个)(核)(心)(的)(问)(题)(。围)(棋)(和)(德)(扑)(代)(表)(了)(两)(类)(完)(全)(不)(同)(的)(游)(戏)(,阿)(尔)(法)(狗)(和)(冷)(扑)(大)(师)(也)(在)(朝)(完)(全)(不)(同)(的)(两)(个)(方)(向)(探)(索)(。  只)(不)(过)(,围)(棋)(在)(“完)(美)(信)(息)(游)(戏)(”中)(属)(于)(高)(难)(度)(水)(平)(,因)(而)(阿)(尔)(法)(狗)(也)(在)(搜)(索)(这)(个)(核)(心)(问)(题)(上)(取)(得)(了)(巅)(峰)(成)(就)(。而)(冷)(扑)(大)(师)(刚)(刚)(打)(开)(了)(“非)(完)(美)(信)(息)(游)(戏)(”的)(大)(门)(,对)(于)(解)(决)(隐)(藏)(信)(息)(这)(个)(问)(题)(来)(说)(,德)(扑)(AI只)(是)(个)(开)(始)(。光)(就)(游)(戏)(而)(言)(,与)(德)(扑)(类)(似)(的)(奥)(马)(哈)(,就)(比)(德)(扑)(更)(为)(复)(杂)(一)(些)(。布)(朗)(说)(,他)(们)(对)(中)(国)(的)(麻)(将)(略)(知)(一)(二)(,麻)(将)(也)(是)(一)(种)(充)(满)(了)(隐)(藏)(信)(息)(的)(游)(戏)(。)!

作者:【龙虎和代理】


相关文章

版权所有:开封开会议费发票 粤ICP备05070829 网站标识码4400008580
主办:南方新闻网 协办:广东省经济和信息化委员会 承办:南方新闻网
建议使用1024×768分辨率 IE7.0以上版本浏览器