Googleã«ããè²·åããã人工ç¥è½éçºãã³ãã£ã¼ã®ãDeepMindããéçºãã人工ç¥è½ã½ããã¦ã§ã¢ã®ãAlphaGoãã¯ãä»ã¾ã§äººå·¥ç¥è½ãåã¦ãªãã£ãå²ç¢ã§ä¸çæå¼·ã®æ£å£«ã§ããã¤ã»ã»ãã«(æä¸ä¹)ä¹æ®µãæã¡ç ´ãã¨ããå¿«æãæãéãã¾ãããéå½ã§è¡ãããå²ç¢çä¸ç´ã®ä¸æ¦ã§ã¯ã人工ç¥è½ã人éãä¸åã£ãã¨ããçµæãåºãããã§ãããDeepMindãçããããã¹ã»ããµãã¹æ°ã¯äººå·¥ç¥è½ã®åå©ã¯å ¨ã¦ã®äººé¡ã«ã¨ã£ã¦å¤§ããªå©çããããã¨ã強調ãã¦ãã¾ãã Official Google Blog: What we learned in Seoul with AlphaGo https://googleblog.blogspot.jp/2016/03/what-we-learned-in-seoul-with-alphago.html 2016å¹´3æ9æ¥ãã3æ15æ¥ã¾ã§è¡ãããAlphaGoã¨ã¤ã»ã»ã
The slides go through the implementation details of Google Deepmind's AlphaGo, a computer Go AI that defeated the European champion. The slides are targeted for beginners in the machine learning area. Korean version (íêµì´ ë²ì ¼): http://www.slideshare.net/ShaneSeungwhanMoon/ss-59226902Read less
Google DeepMind社ãéçºãã¦ããAlpha Goã®è«æãèªãã§ã¿ããã©ããã£ããã¨ããç解ã§ããªããæ©æ¢°å¦ç¿ã®ç¥èã足ããªããstochastic gradient descentã¨stochastic gradient ascentã£ã¦ãªãã ãï¼ãªããã®å¿ 殺æã«ããè¦ããªãããããã調ã¹ãªããèªãã§ã¿ãã¨ãã³ã³ãã¥ã¼ã¿å²ç¢ãåçã«å¼·ããªã£ãåå ã§ããã¢ã³ãã«ã«ãæ¨æ¢ç´¢ã«ããããã¬ã¤ã¢ã¦ãï¼Natureã®è«æã ã¨rolloutï¼æã®æã¡æ¹ï¼Policy networkï¼ã¨ã¢ã³ãã«ã«ãæ¨æ¢ç´¢ã®ãã¼ãã®è©ä¾¡å¤ï¼value networkï¼ã決ããã¨ããã«ç³ã¿è¾¼ã¿ãã¥ã¼ã©ã«ãããã¯ã¼ã¯ã使ã£ã深層å¦ç¿ãé©ç¨ãã¦ããããããã¢ã³ãã«ã«ãæ¨æ¢ç´¢ã«ã¤ãã¦ã¯æ å ±å¦çå¦ä¼å¦ä¼èªãæ å ±å¦çãã®ä»¥ä¸ã®è¨äºãããããããã£ãã ç¾æ·»ä¸æ¨¹: ã¢ã³ãã«ã«ãæ¨æ¢ç´¢-ã³ã³ãã¥ã¼ã¿å²ç¢ã«é©å½ãèµ·ãããæ°ææ³
æä¸ç¦ä¹ã§ããAlphaGoã®ãµã¼ãå©ç¨æéãã¨ã¤ã¤ã§ãè¦ç©ããããã58å9811ä¸4845åã ã£ãããã£ãã60ååãã½ã¼ã¹ã¯Googleæ§ã®GCEãè¦ç©ããã¼ã«ãå¹´é2586ä¸ãã«=29ååããããéçº&å¦ç¿2å¹´ãªã®ã§ãããããããæ ªå¼ä¼ç¤¾ã¾ãã ããããï¼ã¤è²·åãã¦ã10ååãªããªãæ¥ããã©ã¤ã¹ãã¹ããã¯ã¯é©å½ããã ããGPUå©ç¨æéãé¤ããGPUã¤ã³ã¹ã¿ã³ã¹ã¯ä¸è¬ã«å ¬éããã¦ããªãããã ãã ãããæ¬å½ã¯ãã£ã¨é«ããããããªãããCPUãå ¬å¼çºè¡¨ã®1202å°ã§ãã£ã¨ä½¿ã£ãå ´åãªã®ã§åãããªãããã¨ãPersistent Diskãªã©ã¯ï¼åã§è¨å®ãã¦ããã è¨ããããã¨ã¯ãã©ã£ã¡ã«ãã¦ããé¶ç´°ãã³ãã£ã¼ä¼æ¥ã¯æããªãéé¡ã ã¨ãããã¨ãåãã£ãã æéä¼ç¤¾ã·ã¼ãªã¹ã®æ山社é·ã¨è©±ãã¦ãã¦ããAlphaGoã£ã¦Googleã¿ããã«ãµã¼ããGPUãæ»ã¬ã»ã©ä½¿ããªãã¨ç¡çã ãããã«ããããããã
ä»æ¥ã®åè¨ãã¢ã«ãã¡ç¢ã¨ã®æ¦ããè¦ãã¨ã人éã®ç解åãçããããããªãããã天ç¶ç¥è½ããã©ãã«ããªããªããä»å¾ãã¹ã¦ã®æ¥ç¨®ãå ¨ã¦ã®é åã§ããã浸éãã¦ããã人éãåããªãã¦ããæ代ã¾ã§ããç§èªã¿ã ã / â[第ä¸å±] 人工ç¥è½ãâ¦â https://t.co/XyCPK0PT2y â Xevra Lindich (@xev_ra) 2016å¹´3æ13æ¥ å®ã«æç¡éã ã¤ãã«äººå·¥ç¥è½ã人é¡ãè¶ ããäºã確èªããããå¯ä¸äººéãå¾æã ã£ãå²ç¢ã§AIãåã£ãã¨ããäºã¯ãä»å¾AIã¯ã©ããªé åã«ããã¦ãåé¡ããè¨å®ãããã°äººé¡ãæã¡è² ããäºãã§ããäºãæå³ãããããæªæ¥æ°¸å«äººé¡ã¯AIã«åã¤äºã¯ç¡ãã®ã ã ãã¡ãããç¾ç¶ã®AIã¯ä¸ç観ãç¡ãã®ã§ä¾¡å¤è¦³ããªããåé¡ãè¨å®ããªããã ããç´ã¡ã«ç¤¾ä¼ã®ä½ããå¤ãã訳ã§ã¯ãªãããç£æ¥ã®å ¨é åãå ¨åéã§ããããã¨AIã®æ´»ç¨ãé²ã¿ãæ°ãä»ããã人éã«ããããä»äºãç¡ããª
ãªãªã¼ã¹ãé害æ å ±ãªã©ã®ãµã¼ãã¹ã®ãç¥ãã
ææ°ã®äººæ°ã¨ã³ããªã¼ã®é ä¿¡
å¦çãå®è¡ä¸ã§ã
j次ã®ããã¯ãã¼ã¯
kåã®ããã¯ãã¼ã¯
lãã¨ã§èªã
eã³ã¡ã³ãä¸è¦§ãéã
oãã¼ã¸ãéã
{{#tags}}- {{label}}
{{/tags}}