ãªã³ã¯ Qiita Chainerã§é¡ã¤ã©ã¹ãã®èªåçæ - Qiita ![Screen Shot 2015-12-24 at 6.56.28 PM.png](https://qiita-image-store.s3.amazonaws.com/0/51673/11ef48f6-0594-d91c-809... 426 users 14
æ¦è¦ 大äºç«¶é¦¬å ´ã«è¡ãæ©ä¼ããã£ãã®ã§ãæ©æ¢°å¦ç¿ã使ã£ã¦ç«¶é¦¬ã®çµæãäºæ¸¬ã§ãããããã£ã¦ã¿ã¾ããã ãã®çµæãå¸çè³ã§ä¸ä½ãå½ã¦ããã¨ãã§ããã®ã§ãè¨äºãæ¸ãã¾ãã ããªãé©å½ãªäºæ¸¬ãªã®ã§ãéã³ã¨ãã¦è¦ã¦ãããããã¨æãã¾ãã 証æ å½ãã£ãã¨ãã証æ ã«ãè¨å¿µã§ã¨ã£ã馬å¸ç»åã æ©æ¢°å¦ç¿ã§äºæ¸¬ãããã®ã¨ããããã¯ãè¦ã¦äºæ¸¬ãããã®ã¨ãï¼ã¤è²·ãã¾ããã (ã³ã³ã£ã¦è¤åãããã300å) åé¡ã®è¨å® 大äºç«¶é¦¬å ´ã§è¡ãããå¸çè³ã®ï¼ä½ã®ã¿ãå½ã¦ã¾ãã 競馬ã«ã¯ãè²ã ãªé¦¬å¸ã®è²·ãæ¹ãããã¾ãããä»åã¯ç°¡åã§ã·ã³ãã«ãªåé¡è¨å®ã¨ãããã£ãã®ã§ã1ä½ã®ã¿ãäºæ¸¬ãããã¨ã«ãã¾ããã ãã¼ã¿ã®åå¾ æ師ããå¦ç¿ãè¡ãã®ã§ãéå»ã®ç«¶é¦¬çµæã®ãã¼ã¿ãå¿ è¦ã§ãã ãã¡ãã®ãµã¤ããããã¼ã¿ãã¯ãã¼ãªã³ã°ãã¾ããã åé¢æ±4ç«¶é¦¬å ´å ¬å¼ã¦ã§ããµã¤ã ã¬ã¼ã¹æ å ±ã®ãã¼ã¸ãããã¬ã¼ã¹ã«åºã馬ã®éå»æ å ±ããããã¼ã¸ã¸ã®ãªã³
ãã®æ稿ã¯ç±³å½æé 3 æ 26 æ¥ã«æ稿ããããã®ï¼æ稿ã¯ãã¡ãï¼ã®æ訳ã§ãã Posted by Google Cloud ãããããã¼ ã¢ããã±ã¤ã ä½è¤ä¸æ² ãã® 3 ã¤ã®ã©ã¼ã¡ã³ã¯ã41 åºèããã©ã¼ã¡ã³äºéã®ãã¡ 3 åºèã§ä½ããããã®ã§ãããããããã©ã®åºèã§åºããããã®ãåããã¾ããï¼ããã¼ã¿ ãµã¤ã¨ã³ãã£ã¹ãã®åäºè³¢æ²»ãããä½æããæ©æ¢°å¦ç¿ï¼MLï¼ã«ããã©ã¼ã¡ã³èå¥å¨ã使ãã°ãããããã®å¾®å¦ãªçãä»ãã®éããè¦åãããã¨ã§ã95% ã®ç²¾åº¦ã§åºèãç¹å®ã§ãã¾ãã ãã®åçãè¦ã¦ãåããã¨ãããã©ã¼ã¡ã³äºéã®ç¸å½ã³ã¢ãªãã¡ã³ã§ããªããã°ãã©ã¼ã¡ã³ç»åãã 41 åºèã®ã©ãã§ä½ãããããè¦åãããã¨ã¯ç°¡åã§ã¯ããã¾ããããã¼ãã«ãã©ãã¶ãã®è²ãå½¢ã«ãã¾ãéãã®ãªãå ´åãå¤ãã®ã§ãã åäºããã¯ããã£ã¼ã ã©ã¼ãã³ã°ã使ã£ã¦ãã®åé¡ã解ãããèå³ãæã¡ãã¤ã³ã¿ã¼ãããä¸ãã 48
æ¬ããã°ã©ã ã®æ大ã®ç¹å¾´ã®ä¸ã¤ã¯ãå ¨ã¦ã®ãããã¯ã«ã¤ãã¦ãæ¼ç¿ãä¸å¿ã«æ§æããã¦ããç¹ã§ããå®éã«æãåãããªããç解ãé²ãããã¨ã§ãå¹çããå¦ç¿ãããã¨ãã§ãã¾ãã å®éã«ã¢ãã«ãå¦ç¿ãããªããæè¡ãç¿å¾ããæ¬æ ¼çãªæ¼ç¿å 容ã¨ãªã£ã¦ãã¾ããDeep Learningã¯ãã¢ãã«ãå®éã«å¦ç¿ããæ§åã観測ãããã©ã¡ã¼ã¿ã調æ´ãããã¨ã§ã¢ããªã±ã¼ã·ã§ã³ã«å¿ããããã©ã¼ãã³ã¹æ大åãè¡ããã¨ãé常ã«éè¦ãªæè¡ã§ããããã®ä¸é£ã®æµããå ¨ã¦ã®æ¼ç¿ã§çµé¨ããªããéè¦ãªè¦ç´ ã身ã«ã¤ãããã¨ãå¯è½ã§ãã
ãã®è¨äºã¯ç§»è»¢ãã¾ãããç´2ç§å¾ã«æ°è¨äºã¸ç§»åãã¾ãã移åããªãå ´åã¯ã³ã³ãã¯ãªãã¯ãã¦ãã ããã ãããªãã¨ããã¦ã¿ãã âãããããã pythonã«ããæ©æ¢°å¦ç¿ã®åå¼·ãããã®ã§ãå®è·µã¨ãããã¨ã§ã人æ°ã¢ã¤ãã«ãä¹æ¨å46ãã®å人çã«å¥½ããª5人ã®ã¡ã³ãã¼ãåºå¥ãã¦è¦ã¾ããã大ããªæµãã¯ãããªæãã§ãã webä¸ããäºäººã®ç»åã100æãã¤åã£ã¦ãã ç»åããé¡é¨åãåãåºãã¦ä¿åããã¹ããã¼ã¿ã®åãåºã ç»åã®æ°´å¢ã ã¢ãã«ãå®ç¾©ãã¦ãå¦ç¿ ãã¹ã(é¡ãåè§ãå²ã£ã¦ããã®äººã®ååãåºå) 説æã¯ãããªããã«ãã¦ã彼女ãã®å¯æãã«ã¤ãã¦èªãããã¨ããã§ãããããããããã°ã§ã¯ãªãã®ã§ãå°ãæè¡çãªãã¨ãæ¸ãã¾ãã ä»åã¯jupyterã使ã£ã¦ä½æ¥ãé²ãã¾ãããnotebookå½¢å¼ãªã®ã§çµæãè¦ãããåå¿è ã«ã¯ããç°å¢ã§ãããç°å¢ã¯ä»¥ä¸ã macOS:10.13.1 python:3.6.
ã¤ãã«ã¬ãã¢ãæ»ãã å¹´é½¢ãè¶ ãã¦ãã¾ã£ããã¨ã«æ°ãã¤ããèªåãã¾ã ä½ãæãéãã¦ããªããã¨ãæ²ããæãä»æ¥ãã®ããã§ãã ãã¦ãä»æ¥ã¯Googleãåºããæ©æ¢°å¦ç¿ã©ã¤ãã©ãªã®TensorFlowã®ä½¿ãæ¹ã«ã¤ãã¦è»½ã説æãã¤ã¤ãããããã®å¶ä½ä¼ç¤¾ã®èå¥ãè¡ããã¨æãã¾ãã TensorFlowã¨ã¯ TensorFlowã¯Googleã11/9ã«å ¬éããApache 2.0ã©ã¤ã»ã³ã¹ã§ä½¿ããæ©æ¢°å¦ç¿ã©ã¤ãã©ãªã§ããGoogleã¯æ§ã ãªã¨ããã§ãããã¯ãã«æ©æ¢°å¦ç¿ãæ´»ç¨ãã¦ãã¾ãããTensorFlowã¯å®éã«Googleå é¨ã®ç 究ã§ä½¿ããã¦ããããã§ãï¼TensorFlow: Google ææ°ã®æ©æ¢°å¦ç¿ã©ã¤ãã©ãªããªã¼ãã³ã½ã¼ã¹å ¬é - Google Developer Japan Blogï¼ã Googleã®ãã¼ã ããªã¥ã¼ã¯æããããã®ã§ãGitHubã®Staræ°ã¯ãã§ã«ChainerãC
人ã¨AIã®å ±åã主ã¨ãããGoogle PAIRããæ©æ¢°å¦ç¿ï¼NNå«ãï¼ããã©ã¦ã¶ä¸ã§å ¨ã¦å®è¡ã§ããWebMLã©ã¤ãã©ãªãdeeplearn.js 0.1.0ãããªãªã¼ã¹ 2017-08-12 人ãä¸å¿ã¨ããAIã·ã¹ãã ã®ç 究ã¨è¨è¨ãããGoogleã®ããã¸ã§ã¯ããThe People + AI Research Initiativeï¼PAIRï¼ãã¯ããã©ã¦ã¶ä¸ã§æ©æ¢°å¦ç¿ï¼Machine learningï¼ãå ¨ã¦å®è¡ã§ãããªã¼ãã³ã½ã¼ã¹ã©ã¤ãã©ãªãdeeplearn.js 0.1.0ãããªãªã¼ã¹ãã¾ããã deeplearn.js æ¬ã©ã¤ãã©ãªã¯ãå¯è½ãªéãå¤ãã®äººã«æ©æ¢°å¦ç¿ãéããããã¨ããæãã®ãã¨éçºããã¾ããããã®çºãã¤ã³ã¹ãã¼ã«ãããã¯ã¨ã³ããªãããã©ã¦ã¶ä¸ã§å ¨ã¦å®è¡ã§ããä»æ§ã«ãªã£ã¦ãã¾ãã ä»ã¾ã§ãWeb Machine learningï¼WebMLï¼ã©ã¤ãã©ãªã¯åå¨ãã¦ãã¾
以åãããã£ã¦ã¿ããã£ãRaspberry Piã®ç©ä½èªèã試ãã¦ã¿ã¾ãããä»åã¯ã¯ã©ã¦ããµã¼ãã¹ã使ããã«æ·±å±¤å¦ç¿ã©ã¤ãã©ãªã¨å¦ç¿æ¸ã¿ã¢ãã«ã使ã£ã¦ã¿ã¾ããã ç°å¢ Raspberry Pi3 (RASPBIAN JESSIE WITH PIXEL 4.4ã/ Python 3.4.2) LOGICOOL ã¦ã§ãã«ã HDç»è³ª 120ä¸ç»ç´ C270 ããããããPCçç¨ã¹ãã¼ã«ã¼å°åããããç½ 7ã¤ã³ã(1024*600) IPS液æ¶ããã«ããã£ã¹ãã¬ã¤ ä»ã¾ã§ã«ã¡ã©ã¢ã¸ã¥ã¼ã«ãå©ç¨ãã¦ããã®ã§ãããOpenCVã§ã¹ããªã¼ãã³ã°ããããããWebã«ã¡ã©ãè³¼å ¥ãã¾ãããã¤ãã§ã«ã¡ãã£ã¨å¯æãããããããåã®ã¹ãã¼ã«ã¼ãåããã¦è²·ã£ã¦ã¿ã¾ããã ããããåã®ã¹ãã¼ã«ã¼ãWebã«ã¡ã©ã«æ ã£ãç©ãããã¹ã£ã¦ããã¾ãï¼è±èªã§ãï¼ã ãããªæã Deep Learning Object Recog
* ãã®æ稿ã¯ç±³å½æé 5 æ 12 æ¥ã«æ稿ããããã®ï¼æ稿ã¯ãã¡ãï¼ã®æ訳ã§ãã Posted by ä½è¤ä¸æ², Staff Developer Advocate, Google Cloud Cliff Young, Software Engineer, Google Brain David Patterson, Distinguished Engineer, Google Brain Google æ¤ç´¢ãã¹ããªã¼ããã¥ã¼ãGoogle ãã©ããããã¦Google 翻訳ããããã®ãµã¼ãã¹ã«å ±éããã®ã¯ããããããã¥ã¼ã©ã«ãããã¯ã¼ã¯ï¼NNï¼ã®è¨ç®å¦çã®é«éåã®ããã« Google ã®ç¬¬ä¸ä¸ä»£ã® Tensor Processing Unit (TPU) ãç¨ãããã¦ããç¹ã§ãã Google ã® Tensor Processing Unit (TPU) ãæè¼ãããåè·¯åºæ¿ï¼å·¦ï¼ã¨ã G
æ¦è¦ï¼ æ¬ç 究ã§ã¯ï¼ç³ã¿è¾¼ã¿ãã¥ã¼ã©ã«ãããã¯ã¼ã¯ãç¨ãã¦ï¼ã·ã¼ã³ã®å¤§åçãã¤å±æçãªæ´åæ§ãèæ ®ããç»åè£å®ãè¡ãææ³ãææ¡ããï¼ææ¡ããè£å®ãããã¯ã¼ã¯ã¯å ¨å±¤ãç³ã¿è¾¼ã¿å±¤ã§æ§æããï¼ä»»æã®ãµã¤ãºã®ç»åã«ãããèªç±ãªå½¢ç¶ã®ãç©´ããè£å®ã§ããï¼ãã®è£å®ãããã¯ã¼ã¯ã«ï¼ã·ã¼ã³ã®æ´åæ§ãèæ ®ããç»åè£å®ãå¦ç¿ãããããï¼æ¬ç©ã®ç»åã¨è£å®ãããç»åãèå¥ããããã®å¤§åèå¥ãããã¯ã¼ã¯ã¨å±æèå¥ãããã¯ã¼ã¯ãæ§ç¯ããï¼å¤§åèå¥ãããã¯ã¼ã¯ã¯ç»åå ¨ä½ãèªç¶ãªç»åã«ãªã£ã¦ããããè©ä¾¡ãï¼å±æèå¥ãããã¯ã¼ã¯ã¯è£å®é åå¨è¾ºã®ãã詳細ãªæ´åæ§ã«ãã£ã¦ç»åãè©ä¾¡ããï¼ãã®2ã¤ã®èå¥ãããã¯ã¼ã¯ä¸¡æ¹ããã ã¾ããããã«è£å®ãããã¯ã¼ã¯ãå¦ç¿ããããã¨ã§ï¼ã·ã¼ã³å ¨ä½ã§æ´åæ§ãåãã¦ããï¼ãã¤å±æçã«ãèªç¶ãªè£å®ç»åãåºåãããã¨ãã§ããï¼ææ¡ææ³ã«ããï¼æ§ã ãªã·ã¼ã³ã«ããã¦èªç¶ãªç»åè£å®ãå¯è½ã¨ãªãï¼ããã«å¾æ¥ã®
#ã¯ããã« ã¨ãç»åã¯æé«ã§ãã ã¨ãç»åããããã°ã誰ã§ãèªç±ã«ã¨ãããªãã¨ãã§ãã¾ãããã¨ã彼女ãããªãããµã¡ã³ã§ãã£ã¦ããä¸äººã§æ§çã«è奮ãããã¨ãã§ãã¾ããããã«ç¢ºããªæºè¶³ãè¦ãã¦ã幸ãã«æµ¸ããã¨ãã§ãã¾ããã©ããªå好ãæãã®ã¾ã¾ã§ãã æ ã«æã 人é¡ã«ã¨ã£ã¦ãã¨ãç»åãéãããã¨ã¯ããã³ã³ãã¬ã·ãç³ã転ããã®ã¨åãããã«ã種ã¨ãã¦ã®ç¿æ§ã¨ç§°ãã¦ãéè¨ã§ã¯ãªãè¡ããªã®ã§ã¯ãªããã¨æãã¾ãã ããããªãããæã ã¯éé·é¡ã®é·ããçãç©ã§ããããããæ°ä¸å¹´ã«æ¸¡ã£ã¦åãããã«ç³ã転ããç¶ãã¦ãããã³ã³ãã¬ã·ã¨åãã§ã¯ããã¾ãããããå¹ççã«ãããæ欲çã«ã¨ãç»åãåéãã¦ããã®äººé¡ã§ãã ãã ãããã¯è¨ã£ã¦ãã¨ãç»åã®åéã¯é常ã«å¤§å¤ã§ããæ§ã ãªãµã¤ããå·¡ããååã«åå³ããä¸ã§ããã£ããã£ã³ã°ã®ãã£ãåã ãã確ããã¹ãã¼ã ã«åããã¦åéãæ§é åãã¦ããå¿ è¦ãããã®ã§ããæ¥ã«ãã£ã¦å¿ è¦ãªä¸æã
INTRODUCTION ä¸ã®ç»åã¯ï¼2014å¹´ã®ãã¹ãã£ã³ãã¹ç«å½é¤¨ã®ãã®ã§ãï¼ã¿ããªã¨ã¦ãç¾äººã§ããï¼ ãã®ä¸æ¹ã§ï¼ããã¨è¦ãæãï¼ã©ã®æ¹ãåããããªé¡ããã¦ããããã«è¦ãã¾ãï¼é¡ã¯åãå¼ã¶ã®ã§ããããï¼ããã ç«å½é¤¨ã£ã½ãé¡ ã¨å¼ã¶ãã¨ã«ãã¾ãï¼ ã¾ããéå¦ã£ã½ãããå¦ç¿é¢ã«ããããã¿ãããªè¨èãããè³ã«ã¯ãã¿ã¾ããï¼ããããã¯ã éå¦ã£ã½ãé¡ ã å¦ç¿é¢ã£ã½ãé¡ ã¨ãããã®ããããæ ãªããã«æãã¾ãï¼ ããã§ä»åã¯ï¼å¤§å¦ãã¨ã®é¡ã®å¾åã Deep Learning ããï¼ããç¾å¥³ãã©ã®å¤§å¦ã«ãããããå¤å¥ã§ããã¢ãã«ãä½æãã¦ã¿ã¾ããï¼ APPROACH 1. 大å¦ãã¨ã®å¥³æ§ã®ç»ååé ã¾ãï¼å大å¦ã®å¥³æ§ã®ç»åãã²ãããåå¾ãã¾ãï¼ãã¹ã³ã³ãã¹ãã®ãã¼ã¿ã«ãµã¤ã ã«ï¼å大å¦ã®éå»ã®ãã¹ã³ã³ã®åçãä½ç³»çã«ã¾ã¨ã¾ã£ã¦ããã®ã§ï¼å©ç¨ããã¦ããã ãã¾ããï¼ # -*- coding:ut
ã¡ã¾ãã§ã¯ãæ©æ¢°å¦ç¿ããã¼ã ã®ããã§ãã ãããã¾ã£ããæ代ã«ã¤ãã¦ããã¦ãã¾ããã ããããæ©æ¢°å¦ç¿ãç¹ã«èªç¶è¨èªå¦çã«ç²¾éãã人ã®æ¡ç¨ã«ããããä»äºããã¦ããããã ã«ãããããããèªç¶è¨èªå¦çã©ãããæ©æ¢°å¦ç¿ãå ¨ãåãããªãã ããã§ã¯ããããªããã¨ãããã¨ã§ ãæ©æ¢°å¦ç¿ããããªãå¦çã®çããã¨ããµãã£ã¨éè«ãã§ããã¬ãã«ã ãç®æãã¦ã2017å¹´æ£ææãããåå¼·ãå§ãã¾ããã ã¡ãªã¿ã«ãã©ããªã«ããªãæªãã¦ã1æ¥3æéã¾ã§ï¼ã¨æ±ºãã¦ãã¾ãã ããããæ©æ¢°å¦ç¿ã«èå³é¢å¿ãããããã§ã¯ãªã ãããªããã°ãªããªãä»ã®ä»äºããã 家äºè²å ãåªå ãªã®ã§ããããã§ãç¡çããã¨ç¶ããªãããã§ãã ãAIã§ä¸çãå¤ããããï¼ã ã人工ç¥è½ã§æ³åãã§ããªãæªæ¥ããããã ã¿ãããªãæä¸ã®ä¸ãããããªãã»ã©ã®ä½ãã³ã·ã§ã³ã§æ·¡ã ã¨åå¼·ãã¦ãããã㧠éã«ãããããæèä½ã系人éã¯ãããªã«å¤ããªãã§ãã
åºç¤è¬åº§ãæ å½ããNVIDIAã®CUDA & Deep Learning Solution Architectã®æä¸çå¥æ° 2017å¹´1æ17æ¥ã«éå¬ãããNVIDIAã®ãDeep Learning Institute 2017ãã§ã¯ããã£ã¼ãã©ã¼ãã³ã°(深層å¦ç¿)ã®åºç¤è¬åº§ã¨å®éã«NVIDIAã®ãã£ã¼ãã©ã¼ãã³ã°éçºãã¼ã«ã§ãããDIGITSãã使ããã³ãºãªã³ãã¬ã¼ãã³ã°ã»ãã·ã§ã³ãè¡ãããã ãã¥ã¼ã©ã«ãããã¯ã¼ã¯ã®åºç¤ ãã£ã¼ãã©ã¼ãã³ã°ã®åºç¤è¬åº§ã¯ããããããå§ãã人ã®çºã®ãã¨ãããã®ã§ãNVIDIAãã£ã¼ãã©ã¼ãã³ã°é¨ã®æä¸çå¥æ°ããã£ã¼ãã©ã¼ãã³ã°ã®åºæ¬çãªèãæ¹ãç¨èªãªã©ã解説ããã ã¾ãããã£ã¼ãã©ã¼ãã³ã°ã®ä½ç½®ã¥ãã§ãããããã£ã¼ãã©ã¼ãã³ã°ã¯æ©æ¢°å¦ç¿ã®1ã¤ã®æ¹æ³ã§ãããããã¥ã¼ã©ã«ãããã¯ã¼ã¯ã使ãæ©æ¢°å¦ç¿ã®1ã¤ã®ããæ¹ã¨ãããã¨ã«ãªãã
ã½ããã¦ã§ã¢ã¨ã³ã¸ãã¢ãFPGAï¼field-programmable gate arrayï¼ã使ããã¼ãã«ãããã«ä¸ãã£ã¦ãã¦ãããã¯ã©ã¦ããµã¼ãã¹ã§FPGAãæ´»ç¨ã§ããããPythonã§è¨è¿°ãããã¥ã¼ã©ã«ãããã¯ã¼ã¯ãFPGAã«é«ä½åæã§ããç 究ææãåºã¦ããããã¦ããã®ã ã ã½ããã¦ã§ã¢éçºè ã®ç«å ´ã§FPGAã«åãçµãã¤ãã³ããFPGAã¨ã¯ã¹ããªã¼ã ã»ã³ã³ãã¥ã¼ãã£ã³ã°ãã主宰ããä½è¤ä¸æ²æ°ãFPGAã®é«ä½åæã«ãããã£ã¼ãã©ã¼ãã³ã°ã«ã¤ãã¦ç 究ãã¦ããæ±äº¬å·¥æ¥å¤§å¦ã®ä¸ååè²´æ°ï¼ä¸åç 究室ï¼ãããã¦FPGAãã³ãã¼ã§ããã¶ã¤ãªã³ã¯ã¹ã®ç¥ä¿ç´å¼æ°ããæ¥æ¿ã«å¸¸èãå¤ããã¤ã¤ããFPGAã®ååãèªãåã£ãã æ¬ç¨¿ã§ã¯åº§è«ä¼ã®ä¸ãããã½ããã¦ã§ã¢ã¨ã³ã¸ãã¢ã«FPGAãé«ä½åæãæ±ããããç¾ç¶ãããã¦ãä»å¾ã©ã®ãããªãã¼ã«ã使ãã¹ãããã½ããã¦ã§ã¢ã¨ã³ã¸ãã¢ãFPGAã«åãçµãéã®èª²é¡ãªã©ã«ã¤
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}}