æ©æ¢°å¦ç¿ã®ã¨ãã»ã³ã¹ -å®è£ ããªããå¦ã¶Python,æ°å¦,ã¢ã«ã´ãªãºã - (Machine Learning) ä½è : å è¤å ¬ä¸åºç社/ã¡ã¼ã«ã¼: SBã¯ãªã¨ã¤ãã£ãçºå£²æ¥: 2018/09/21ã¡ãã£ã¢: åè¡æ¬ãã®ååãå«ãããã°ãè¦ãçºå£²ããã¦ããã ãã¶çµã¡ã¾ãããæ§æ³æ®µéã®é ããèè ã®ãã¯ãããããããã¨å è¤å ¬ä¸ããããã話ã伺ã£ã¦ãã¦æ³¨ç®ãã¦ãããã¡ãã®ä¸åãããããä¸éãèªã¿ã¾ããã®ã§ããµã¯ãã¨æ¸è©ãããä½ããæ¸ãã¦ã¿ãããã¨æãã¾ãã åç« ã®æ¦è¦ è¨ãã¾ã§ããªãå®éã®å 容ã¯çæ§ãèªèº«ã§ãèªã¿ããã ãããã®ã§ãããããã¾ã§ã®æ¸è©è¨äºåæ§ã«æ¦è¦ãç°¡åã«ã¾ã¨ãã¦ããã¾ãã 第01ç« ãå¦ç¿ãå§ããåã« Pythonç°å¢ãAnacondaã®ã¤ã³ã¹ãã¼ã«ã«ã¤ãã¦ã®èª¬æããªããã¦ãããã§ãããéè¦ãªã®ã¯å¾è¿°ãããæ¬æ¸ã¯ä½ãå«ã¾ãªãããã¨ããç¯ãããã«æ¬æ¸ã®çãã®å ¨ã¦ãæ¸ããã¦ããã¨è¨ã£ã¦
è±èªçã¯ãã¡ãã TensorFlowã®ç»å ´ä»¥éãOSSãã¼ã¹ã®æ©æ¢°å¦ç¿ã®çãä¸ããã¯å éãã¦ãã¾ããKerasã®ä½è ã®François Cholletããã®è¨èãããã®ç¶æ³ãé常ã«ç«¯çã«è¡¨ãã¦ãã¾ããããã ãã§ãååã ã¨ã¯æãã¾ããããã®è¨äºã§ã¯ããªããªã¼ãã³ã½ã¼ã¹ã®æ©æ¢°å¦ç¿ãå¼·ãã®ããæè¿ã®ã©ããã£ãæµããããã®ããæ´çãããã¨æãã¾ãã tl;dræ©æ¢°å¦ç¿ãDeep Learningã®ãã¬ã¼ã ã¯ã¼ã¯ãå å®ãã¦ããè«æãæ»èªåã«å ¬éãããä»ç¤¾ãç°¡åã«ã¢ã«ã´ãªãºã ã®æ¤è¨¼ãã§ããããã«ãªã£ãå¤ãã®ãã¬ã¼ã¤ã¼ã®åæ¦ã«ãããã¢ã«ããã¢ã§ã®æ©æ¢°å¦ç¿ã®ç 究ãã¬ãããªã¼ã·ã£ã³åããä»ç¤¾ã«ãªãã¢ã«ã´ãªãºã ã§ä¸çºåè² ãå®è£ ã¯ç§å¯ãã¨ããã¢ããã¼ããå³ããç§æçãªæ代5å¹´å10å¹´åã®ä¸çã§ã¯ãå 端ã®æ©æ¢°å¦ç¿ã«åãçµãã§ããã®ã¯å¤§å¦ãªã©ã®ç 究室ã大ä¼æ¥ã®ç 究æãä¸é¨ã®å é²çãªä¼æ¥ãã»ã¨ãã©ã§ãããç¹ã«ãã©ã
ãã¹ã²ã¼ããããä»ã®æ¥æ¬ã®æè¡ãâ¦â¦ã ãä¸éã¯ããã¾ã§é²æ©ãã¦ããã®ãã éçºããã®ã¯ãå µåº«ç西èå¸ã«æ¬ç¤¾ãç½®ãã·ã¹ãã éçºä¼ç¤¾ã»ãã¬ã¤ã³ãåµæ¥35å¹´ããã¾ã社å¡20人ã®ãã¡ç´16人ãã¨ã³ã¸ãã¢ã¨ãããçç²ã®æè¡è éå£ã ã ç´10å¹´åã«ã¼ãããéçºã¹ã¿ã¼ã ãã·ã³ã®ååã¯ãBakeryScanãï¼ãã¼ã«ãªã¼ã¹ãã£ã³ï¼ãããåºã«æä¾ãå§ããã®ã¯ä»ãã4å¹´ã»ã©åãæè¿ã«ãªã£ã¦çªç¶ããããã§ãããåé¿ããããã¨äººã«è¨ããã¦é©ãããââãã¬ã¤ã³ã®åé²ä¹ä»å·è¡å½¹å¡ã¯ãã話ãã BakeryScanã®éçºãå§ã¾ã£ãã®ã¯2008å¹´ã«ããã®ã¼ãããã£ããã¯ãå°å ã»å µåº«çã®ãã³åºç¤¾é·ããç¸è«ãåãããã¨ã ã£ãã ã人ã足ããªãã¦å°ã£ã¦ãããçµé¨ã®æµ ãå¤å½äººã¹ã¿ããã§ãã¬ã¸æã¡ãæ¥å®¢ãã§ãããããªã·ã¹ãã ãä½ã£ã¦ã»ãããââã ã ããå社ã®ãã³ã«é¢ããå°éç¥èã¯ã¼ããããããå¾ ã¡åãã¦ããã®ã¯ãç´6å¹´ã«ãã
ããã«ã¡ã¯ãã¹ãã¼ããã¥ã¼ã¹ã®å¾³æ°¸ã§ãã深層å¦ç¿æ¥çã¯GANã ã¨ã深層強åå¦ç¿ã ã¨ãã§çãä¸ãã£ã¦ãã¾ãããä»æ¥ã¯æ·¡ã ã¨ããã¥ã¼ã©ã«ãããã¯ã¼ã¯ã®éååã®è©±ããã¾ãã TL;DR ãã©ã¡ã¼ã¿ã ããéååããã®ã§ããã°ãã»ã¼ç²¾åº¦ãè½ã¨ããã«ããã©ã¡ã¼ã¿ã®ãã¼ã¿å®¹éã¯1/16ç¨åº¦ã«ã¾ã§åæ¸ã§ãã ãã©ã¡ã¼ã¿ãã¢ã¯ãã£ãã¼ã·ã§ã³ãå¾é ã®ãã¹ã¦ãéååããæ¨è«ã ãã§ãªãå¦ç¿ã¾ã§ããããæ¼ç®ã ãã§å®ç¾ããç 究ãé²ãã§ãã¦ãã ç¾å¨ã¯æ·±å±¤å¦ç¿ = GPUå¿ é ã¨ãããããã®å¢ãããããã©ãéååã®ç 究ãé²ãã¨ãä»å¾ã¯ã©ããªããããããªãã ã¯ããã« æ å ±çè«ã«ãããéååã¨ã¯ãã¢ããã°ãªéãé¢æ£çãªå¤ã§è¿ä¼¼çã«è¡¨ç¾ãããã¨ãæãã¾ãããæ¬ç¨¿ã«ãããéååã¯å³å¯ã«è¨ãã¨ã¡ãã£ã¨æå³ãéããååãªï¼=32bitãããã¯16bitï¼ç²¾åº¦ã§è¡¨ç¾ããã¦ããéãããã£ã¨å°ãªããããæ°ã§è¡¨ç¾ãããã¨ãè¨ãã¾ãã ãã¥
ãã®è¨äºã¯ãã¬ã¿ Advent Calendar 2016ã®22æ¥ç®ã§ãã 21æ¥ç®ã¯swdhã® ActiveRecordãªãã¸ã§ã¯ããé¢é£ãã¨ã·ãªã¢ã©ã¤ãºãã¦ãã·ãªã¢ã©ã¤ãºããã§ããã ã¹ãããã·ã§ããçã«ãã®æç¹ã®ã¢ãã«ãé¢é£ã¢ãã«å«ãã¦ä¿åããããã£ã¦ããè¦æã¯BtoBãã£ã¦ãã¨çµæ§ééãã¾ããããã¼ãã«ãã¡ããã¨æ£è¦åããã°ããã»ã©é£ãããªããã¤ãªã®ã§gemåããã¦ãã¨ãããããã§ãã ãã¦ããã®è¨äºã§ã¯ã¼ãããä½ãDeep Learning âPythonã§å¦ã¶ãã£ã¼ãã©ã¼ãã³ã°ã®çè«ã¨å®è£ ãèªãã§pythonã«å ¥éããã¨ããããåãã¦ãã¥ã¼ã©ã«ãããã¯ã¼ã¯ãå®éã«å®è£ ãã¦è¦ãææãè¨è¿°ãã¾ããå¹³ããè¨ãã°èªæ¸ææ³æã§ãã ã¼ãããä½ãDeep Learning âPythonã§å¦ã¶ãã£ã¼ãã©ã¼ãã³ã°ã®çè«ã¨å®è£ ä½è : æè¤åº·æ¯ åºç社/ã¡ã¼ã«ã¼: ãªã©ã¤ãªã¼ã¸ã£ãã³çºå£²æ¥: 2
2018å¹´4æ25æ¥ããã¡ã¾ãã¦ã ãCodeIQãã®ããã°ã©ãã³ã°è 試ããµã¼ãã¹ãå¹´å確ç´ã¹ã«ã¦ããµã¼ãã¹ã¯ã ITã¨ã³ã¸ãã¢ã®ããã®å¹´å確ç´ã¹ã«ã¦ããµã¼ãã¹ãmoffers by CodeIQãhttps://moffers.jp/ ã¸ä¸æ¬åãããã¾ããã ããã¾ã§å¤ãã®ITã¨ã³ã¸ãã¢ã®æ¹ã«ãCodeIQãããå©ç¨ããã ãã¾ãã¦ã æ¹ãã¦å¿ããæ·±ã御礼ç³ãä¸ãã¾ãã ã¾ããã¨ã³ã¸ãã¢ã®ããã®Webãã¬ã¸ã³ãCodeIQ MAGAZINEãã¯ã ãªã¯ããNEXTã¸ã£ã¼ãã«( https://next.rikunabi.com/journal/ )ã«ä¸é¨ã®è¨äºã®ç§»è¡ãäºå®ãã¦ããã¾ãã ä»å¾ã¯ãmoffers by CodeIQãã«ã¦ã ITã¨ã³ã¸ãã¢ã®çæ§ã®ããè¯ã転è·ããµãã¼ãããããã«ãããä¸å±¤åªãã¦ã¾ããã¾ãã®ã§ã å¼ãç¶ããæ顧ã®ã»ã©ä½åãããããé¡ãç³ãä¸ãã¾ãã ã¾ããCod
ããã¯ãæ©æ¢°å¦ç¿ã«é¢ããåºç¤ç¥èãã¾ã¨ããã·ãªã¼ãºè¨äºã®ç®æ¬¡ã¨ãªãè¨äºã§ããã¾ã¨ãããã¨ã§ç¥èãä½ç³»åã§ãã¦èªåèªèº«ã®çºã«ããªãã®ã§ãããããã¢ã¦ããããããããã¨ã¯å¤§äºã ã¨æã£ã¦ãã¾ãããã ãæ®éã«ããã°è¨äºãæ¸ãã®ãé¢ç½ããªãã®ã§ãã¡ãã£ã¨ãã¤ãã¨ã¯éãæ¹æ³ã§ãã£ã¦ã¿ããã¨ããã®ãä»åã®ã·ãªã¼ãºè¨äºã 2 ã¡ããããã®ãã£ã©ã¯ã¿ã¼ãç»å ´äººç©ã¨ãã¦åºã¦ãã¦ãå½¼ããä¼è©±ãã¦è©±ãé²ãã§ããããã夫ã§å¦ã¶ã·ãªã¼ãºãã¨ããè¬ç¾©èª¿ã®å½¢å¼ã®ãã®ãããã¾ããå人çã«ã¯ãã夫ã§å¦ã¶ã·ãªã¼ãºã æ°å¦ã¬ã¼ã« ã®ãããªä¼è©±å½¢å¼ã§è©±ãé²ãã§ããèªã¿ç©ã¯èªã¿ãããã¨æã£ã¦ãã¾ããããã«ãå æ¥ã¿ã¤ãã ãã夫ã§å¦ã¶ãã£ã¸ã¿ã«ä¿¡å·å¦ç ã¨ããè³æãã¨ã¦ã¤ããªããããããããããã®çä¼¼ããã¦æ¸ãã¦ã¿ããã¨æãè³ãã¾ãããè¨äºä¸ã®ãã夫ã¨ãããªã夫ã®ã¢ã¤ã³ã³ã¯ http://matsucon.net/material/m
移転ãã¾ããã https://chezo.uno/post/2016-05-29-sonomoderu-guo-xue-xi-siteruno-wei-xue-xi-nano-tokun-tutara/
2. ç®æ¬¡ â¢â¯ Deep Learning ã¨ã¯" â⯠æ©æ¢°å¦ç¿ã«ã¤ãã¦" â⯠å¾æ¥ã® NN ã¨ã®ã¡ãã" â⯠Deep Learning ã®ãã¬ã¤ã¯ã¹ã«ã¼" â¢â¯ dA (Denoising Autoencoders) ããããã" â⯠æ°çã¢ãã«ã®è§£èª¬" â⯠Python ã§å®è£ ããåæºå" â⯠ã³ã¼ãã¬ãã¥ã¼" â⯠å®è¡çµæ" â¢â¯ RBM (Restricted Boltzmann Machines) ããããã" â⯠æ°çã¢ãã«ã®è§£èª¬" â⯠å®è¡çµæ" â¢â¯ ã¾ã¨ã 4. Deep Learning ã¨ã¯ â¢â¯ å ¥åä¿¡å·ããããæ½è±¡çãªæ¦å¿µãå¦ã¶ã»ç¹å¾´ãæ½åºãã æ©æ¢°å¦ç¿ã®ææ³ã®éåã§ã " âãã¥ã¼ã©ã«ãããã¨ã©ãéãã®ï¼â! â¢â¯ ãã¥ã¼ã©ã«ããããå¤å±¤ã«ãããã§ã " âå¾æ¥ã®ãã¥ã¼ã©ã«ãããã¯ã¼ã¯ã¨ä½ãéãã®ï¼â! â¢â¯ ã²ã¨ã¤ã²ã¨ã¤ã®ã¬ã¤ã¤ã¼éã§ãã©
ã©ã³ãã³ã°
ã©ã³ãã³ã°
ã©ã³ãã³ã°
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}}