TPUè«–æ–‡ã®ç¿»è¨³ï¼ˆï¼‘）
データセンター内ã§ã® Tensor Processing Unitã®ãƒ‘ãƒ•ã‚©ãƒ¼ãƒžãƒ³ã‚¹è§£æž Norman P. Jouppi, Cliff Young, Nishant Patil, David Patterson, Gaurav Agrawal, Raminder Bajwa, Sarah Bates, Suresh Bhatia, Nan Boden, Al Borchers, Rick Boyle, Pierre-luc Cantin, Clifford Chao, Chris Clark, Jeremy Coriell, Mike Daley, Matt Dau, Jeffrey Dean, Ben Gelb, Tara Vazir Ghaemmaghami, Rajendra Gottipati, William Gulland, Robert Hagmann, C. Richard
Using Cloud Dataflow to run parallel predictions with your TensorFlow model - ã‚ã‚‚ã‚ã‚‚
Suppose that you've finished training your prediction model with TensorFlow, yeay! Now you have to make predictions with the trained model for tens of thousands of data. How will you do it? I will show you one of possible choices. You can use Cloud Dataflow for general parallel batch processing and it's not hard to apply machine learning models in this framework. What is Cloud Dataflow? Cloud Data
TensorFlow machine learning with financial data on Google Cloud Platform | Google Cloud Blog
TensorFlow machine learning with financial data on Google Cloud Platform If you knew what happened in the London markets, how accurately could you predict what will happen in New York? It turns out, this is a great scenario to be tackled by machine learning! The premise for this problem is that by following the sun and using data from markets that close earlier, such as London that closes 4.5 hour
Distributed TensorFlowã®è©± - Qiita
Deleted articles cannot be recovered. Draft of this article would be also deleted. Are you sure you want to delete this article? Hadoop Conference Japan 2016 ã‚‚ã¨ã‚‚ã¨ã¯2月8æ—¥ã«é–‹å‚¬ã•ã‚Œã‚‹Hadoop Conference Japan 2016ã®ã‚»ãƒƒã‚·ãƒ§ãƒ³ã¨ã—ã¦ã“ã®è©±ã‚’応募ã—ãŸã®ã§ã™ãŒã€ã‚ãˆãªãè½é¸ã—ã¾ã—ãŸâ€¦â€¦ï¼ˆ;_;) ã—ã‹ã—ã€ã‚ã‚ŠãŒãŸã„ã“ã¨ã«å¾©æ´»æˆ¦ã®Lightning Talkã®æŠ•ç¥¨ã§ã¯5ä½ã«é¸ã‚“ã§ã„ãŸã ãã€ãƒ©ãƒ³ãƒã‚¿ã‚¤ãƒ Aä¼šå ´ã§ãŠè©±ã™ã‚‹ã“ã¨ã«ãªã‚Šã¾ã—ãŸã€‚ã‚ã‚ŠãŒã¨ã†ã”ã–ã„ã¾ã™ï¼ 今回ã®ã‚¹ãƒ©ã‚¤ãƒ‰ã¯ã“ã“ã§å…¬é–‹ã—ã¦ã„ã¾ã™ã€‚ ã¨ã¯ã„ãˆã€5分ã®LTã§ã¯ã“ã®å†…容をã—ã£ã‹ã‚Šã¨ä¼ãˆã‚‰ã‚Œã‚‹è‡ªä¿¡ãŒãªã„ã®ã§ã€ä»¥ä¸‹ã§ã‚¹ãƒ©ã‚¤ãƒ‰å†…容ã®è©³ã—ã„解説をã—ãŸã„ã¨æ€ã„ã¾
Cloud Vision APIã®å‡„ã•ã‚’ä¼ãˆã‚‹ã¹ãRasPi botã¨ãƒ“デオを作ã£ãŸè©±
（ã“ã®è¨˜äº‹ã¯Google Cloud Platform Advent Calendar 2015ã®12月3日分ã®è¨˜äº‹ã§ã™ï¼‰ Cloud Vision APIã¨ç§ Googleã«å…¥ã£ã¦ã‹ã‚‰ã¾ã‚‚ãªã5å¹´ã€Google Cloud Platformã®ãƒ‡ãƒ™ãƒãƒƒãƒ‘ーアドボケイト（エãƒãƒ³ã‚¸ã‚§ãƒªã‚¹ãƒˆã¿ãŸã„ãªå½¹å‰²ï¼‰ã®ä»•äº‹ã«å°±ã„ã¦ã‹ã‚‰1å¹´ãŒçµŒã¡ã¾ã—ãŸã€‚仕事ã®åŠåˆ†ã¯ã‚¢ã‚¸ã‚¢åœ°åŸŸå‘ã‘ã®é–‹ç™ºè€…コミュニティ支æ´ã§ã€æ®‹ã‚ŠåŠåˆ†ã¯GCPã®æ–°è£½å“ãƒãƒ¼ãƒ³ãƒã®æ”¯æ´ã‚’ã‚°ãƒãƒ¼ãƒãƒ«å‘ã‘ã«è¡Œã£ã¦ã„ã¾ã™ã€‚ 特ã«ã“ã“åŠå¹´ã¯ã€TensorFlowã‚’ã¯ã˜ã‚ã€GCPã®æ©Ÿæ¢°å¦ç¿’系プãƒãƒ€ã‚¯ãƒˆã®ãƒãƒ¼ãƒ³ãƒæ”¯æ´ã«ãƒ•ã‚©ãƒ¼ã‚«ã‚¹ã—ã¦ã„ã¾ã™ã€‚TensorFlowã¯ãã®åºç« ã§ã€å…¬é–‹å‰ã‹ã‚‰Alphaカスタマーå‘ã‘ã®ã‚¹ãƒ©ã‚¤ãƒ‰ã‚’作ã£ãŸã‚Šèª¬æ˜Žã‚„デモã—ãŸã‚Šã—ã¦ã„ã¾ã—ãŸã€‚ ãã†ã—ãŸGCPã®æ–°ã—ã„機械å¦ç¿’系サービスã®ã²ã¨ã¤ãŒã€Cloud Vision APIã§ã™ã€‚ã“ã‚Œã¯Googl
TensorFlow Paper 感想
当方機械å¦ç¿’ç´ äººã«ã¤ã大ã—ã¦èˆˆå‘³ã¯ãªã‹ã£ãŸã‚‚ã®ã®ã€å®Ÿã¯ Jeff Dean 案件ã ã¨æ°—付ã whitepaper ãらã„ã¯èªã‚€ã“ã¨ã«ã—ãŸã€‚ ãŽã‚‚ã‚“: Jeff Dean ã¨ã„ãˆã° MapReduce ã‚„ GFS を作ã£ãŸ Google ã®ç¥žè©±ç´šãƒ—ãƒã‚°ãƒ©ãƒžã€‚ãã‚“ãªåˆ†æ•£ã‚¤ãƒ³ãƒ•ãƒ©ã®é”人ãŒãªãœã¾ãŸæ·±å±¤å¦ç¿’ã«æ‰‹ã‚’出ã—ãŸã®ã ã‚ã†ã€‚ ã‚ã‹ã£ãŸã“ã¨: TensorFlow ã¯ã€è¡Œåˆ—(ã¨ã„ã†ã‹ãƒ†ãƒ³ã‚½ãƒ«)ã«ç‰¹åŒ–ã—ãŸãƒ‡ãƒ¼ã‚¿ãƒ•ãƒãƒ¼ãƒ»ãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚°ã®åˆ†æ•£å®Ÿè¡Œå‡¦ç†ç³»ã ã£ãŸã€‚ データフãƒãƒ¼ãƒ»ãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚°ã¨ã¯ã€ãƒ‡ãƒ¼ã‚¿ã‚’å—ã‘å–り何ã‹è¨ˆç®—ã—ã¦çµæžœã‚’誰ã‹ã«æ¸¡ã™ã€ã¨ã„ã†å˜ä½ã®ã‚ªãƒ–ジェクト(ノードã€ã‚«ãƒ¼ãƒãƒ«ãªã©ã¨å‘¼ã¶)ã‚’ã¤ãªãŽåˆã‚ã›ã¦ã‚°ãƒ©ãƒ•ã‚’ã¤ãã‚Šã€ã‚ˆã‚Šå¤§ããªè¨ˆç®—を表ç¾ã™ã‚‹æŠ½è±¡åŒ–ã®ãƒ‘ターン。最近ã¯ãƒªã‚¢ã‚¯ãƒ†ã‚£ãƒ–ã®æ–‡è„ˆã§ç›®ã«ã™ã‚‹ã“ã¨ãŒå¢—ãˆãŸã€‚ ãã—㦠MapReduce/Hadoop も今ã¯ãƒ‡ãƒ¼ã‚¿ãƒ•ãƒãƒ¼ã®æž 組ã¿ã§ã‚³ãƒ¼ãƒ‰ã‚’書ãã“ã¨ãŒå¤š
複åˆæ©Ÿæ¢°å¦ç¿’ã«ã‚ˆã‚‹åˆ†æ•£ãƒ‡ãƒ¼ã‚¿å‡¦ç†ã§ã•ã™ãŒã«è‰¯ã出æ¥ã¦ã‚‹TensorFlow
全脳アーã‚テクãƒãƒ£ãƒ»ã‚¤ãƒ‹ã‚·ã‚¢ãƒ†ã‚£ãƒ–副代表ã®é«˜æ©‹ã§ã™ã€‚ å··ã§è©±é¡Œã®TensorFlow(TF)ã®white paperã¨ã‚½ãƒ¼ã‚¹è¦‹ã¾ã—ãŸã€‚ソースも見ã¾ã—ãŸãŒã€whitepaperãŒã¨ã¦ã‚‚よã書ã‘ã¦ã‚‹ã®ã§ã“ã‚Œã ã‘ã§å¤§ä½“è¨è¨ˆãŒç†è§£ã§ãるよã†ã«ãªã£ã¦ã¾ã™ã€‚ ã§ã€çµè«–ã‹ã‚‰ã§ã™ãŒã€ã“ã‚Œã•ã™ãŒã¨ã„ã†ã‹ã™ã”ãよã出æ¥ã¦ã¾ã™ã€‚ ãŸã ã€ãƒãƒœãƒƒãƒˆãƒŸãƒ‰ãƒ«ã‚¦ã‚§ã‚¢ã®ROSãªã©ã‚ˆã‚Šã‹ã¯ãƒ‡ãƒ¼ã‚¿è§£æžã‚½ãƒ•ãƒˆã‚¦ã‚¨ã‚¢ã«è¿‘ãã€ã“ã‚Œã¨æ¯”ã¹ã‚‹ã¨BriCAã¯ã‹ãªã‚ŠROSã«è¿‘ã„ã§ã™ã€‚ 何ãŒé•ã†ã‹ã§ã™ãŒï¼›ã€€TFã¯å¤šæ¬¡å…ƒé…列（テンソル）ã«å¯¾ã™ã‚‹æ“作を定義ã™ã‚‹ã‚«ãƒ¼ãƒãƒ«ï¼ˆæ£ç¢ºã«è¨€ã†ã¨æŠ½è±¡æ“作ã®operationã«ãƒ‘ラメータを与ãˆã¦å…·ä½“çš„ãªè¨ˆç®—æ‰‹é †ã«ã—ãŸã®ãŒã‚«ãƒ¼ãƒãƒ«ï¼‰ã‚’有å‘グラフã§è¡¨ã—ã¦ã€ã‚°ãƒ©ãƒ•å…¨ä½“ã§ãªã‚“らã‹ã®ãƒ‡ãƒ¼ã‚¿å‡¦ç†ãƒ‘イプラインを表ç¾ã€å®Ÿè¡Œã—ã¾ã™ã€‚ ã“ã®å ´åˆã®ã‚°ãƒ©ãƒ•ã¯BriCAã‚„ROSã§ã„ã†ã¨ã“ã‚ã®ãƒŽãƒ¼ãƒ‰ã‚„モジュールをçµåˆã—ãŸèªçŸ¥ã‚¢ãƒ¼ã‚ã‚„ãƒãƒœãƒƒãƒˆ
TensorFlow 畳ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã§æ‰‹æ›¸ãèªè˜çŽ‡99.2%ã®åˆ†é¡žå™¨ã‚’構築 - Qiita
TensorFlowã¨ã¯2015/11/9ã«ã‚ªãƒ¼ãƒ—ンソース化ã•ã‚ŒãŸGoogleã®æ©Ÿæ¢°å¦ç¿’ライブラリã§ã™ã€‚ã“ã®è¨˜äº‹ã§ã¯ãƒ‡ã‚£ãƒ¼ãƒ—ラーニングã¨è¨€ã‚ã‚Œã‚‹å¤šå±¤æ§‹é€ ã®ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚’TensorFlowを利用ã—ã¦æ§‹ç¯‰ã—ã¦ã„ã¾ã™ã€‚ TensorFlowã¯Pythonã‹ã‚‰æ“作ã§ãã¾ã™ãŒãƒãƒƒã‚¯ã‚¨ãƒ³ãƒ‰ã§ã¯C++ã§é«˜é€Ÿã«è¨ˆç®—ã—ã¦ã„ã¾ã™ã€‚macã®Python2.7系環境ã§TensorFlowã®ä¸Šç´šè€…用ãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ«ã‚’è¡Œã„ã€æ‰‹æ›¸ãèªè˜çŽ‡99.2%ã®å¤šå±¤æ§‹é€ ã®ç•³ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ãƒ¢ãƒ‡ãƒ«ã®åˆ†é¡žå™¨ã‚’構築ã—ãŸã¨ãã®ä½œæ¥ãƒ¡ãƒ¢ã§ã™ã€‚特別ãªè¨å®šãªã—ã«CPU使用率270%メモリ600MByteã¨ã¡ã‚ƒã‚“ã¨ä¸¦åˆ—計算ã—ã¦ãã‚Œã¾ã—ãŸã€‚MNISTランã‚ングを見るã¨èªè˜çŽ‡99.2%ã¯ä¸Šä½ã®ãƒ¢ãƒ‡ãƒ«ã¨ãªã‚‹ã‚ˆã†ã§ã™ã€‚ TensorFlowãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ« TensorFlowã®åˆå¿ƒè€…用ã¨ä¸Šç´šè€…用ãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ«2ã¤ã«å–り組んã§ã¿ã¾ã—ãŸ
1
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
tensorflow/tensorflow/core/distributed_runtime/README.md at master · tensorflow/tensorflow
{{#tags}}- {{label}}
{{/tags}}