imported>Ohtashinichiro：/* 外部链接 */

2026-03-30T08:34:51Z

外部链接

新页面

{{术语}}
{{TA|G1=IT}}
{{Infobox software
|name=TensorFlow
|logo=File:TensorFlow logo.svg
|logo size=260px
|developer=[[谷歌大脑]]团队<ref name=Credits />
|released={{start date and age|2015|11|09}}
|programming language=[[Python]]、[[C++]]、[[CUDA]]
|platform=[[Linux]]、[[macOS]]、[[Windows]]、[[Android]]
|genre=[[机器学习]][[库 (计算机)|库]]
|license=[[Apache 2.0|Apache 2.0开源许可证]]
|website={{URL|https://www.tensorflow.org/}}<br>{{URL|https://tensorflow.google.cn/}}<ref>{{Cite web |url=https://studygolang.com/topics/3393 |title=Google 为 TensorFlow 启用 tensorflow.google.cn 域名 |accessdate=2020-11-08 |archive-date=2020-11-16 |archive-url=https://web.archive.org/web/20201116054254/https://studygolang.com/topics/3393 |dead-url=no }}</ref>
}}
'''TensorFlow'''是一个[[开放源代码|开源]][[库 (计算机)|软件库]]，用于各种感知和语言理解任务的[[机器学习]]。<ref name="YoutubeClip">[https://www.youtube.com/watch?v=oZikw5k_2FM "TensorFlow: Open source machine learning"] {{Wayback|url=https://www.youtube.com/watch?v=oZikw5k_2FM |date=20161215203539 }} "It is machine learning software being used for various kinds of perceptual and language understanding tasks" — Jeffrey Dean, minute 0:47 / 2:17 from Youtube clip</ref>目前广泛地用于研究和生产中，比如[[Google产品列表|Google商业产品]]<ref name="whitepaper2015">{{Cite web|url=http://download.tensorflow.org/paper/whitepaper2015.pdf|title=TensorFlow: Large-scale machine learning on heterogeneous systems|accessdate=2015-11-10|date=2015-11-09|publisher=Google Research|archive-url=https://web.archive.org/web/20151120004649/http://download.tensorflow.org/paper/whitepaper2015.pdf|archive-date=2015-11-20|dead-url=no}}</ref>{{Rp|p.2}}，如[[语音辨識]]、[[Gmail]]、[[Google 相册]]和搜索<ref name="YoutubeClip" />{{Rp|0:26/2:17}}，其中许多产品曾使用过其前任软件DistBelief。

TensorFlow最初由[[谷歌大脑]]团队开发，用于[[Google]]的研究和生产，于2015年11月9日在[[Apache 2.0]]开源许可证下发布。<ref name="Credits">{{Cite web|url=http://tensorflow.org/about|title=Credits|accessdate=2015-11-10|archive-url=https://web.archive.org/web/20151117032147/http://tensorflow.org/about|archive-date=2015-11-17|dead-url=no}}</ref><ref name="Metz-Nov9">{{Cite web|url=http://www.wired.com/2015/11/google-open-sources-its-artificial-intelligence-engine/|title=Google Just Open Sourced TensorFlow, Its Artificial Intelligence Engine|accessdate=2015-11-10|date=2015-11-09|archive-url=https://web.archive.org/web/20151110064825/http://www.wired.com/2015/11/google-open-sources-its-artificial-intelligence-engine/|archive-date=2015-11-10|dead-url=no}}</ref>

== 历史 ==

=== DistBelief ===
从2010年开始，谷歌大脑建立DistBelief作为他们的第一代[[专有]]的[[机器学习]]系统。50多个团队在Google和其他[[Alphabet]]公司在商业产品部署了DistBelief的[[深度学习]][[人工神经网络|神经网络]]，包括[[Google搜索]]、[[Google语音搜索]]、广告、[[Google 相册]]、[[Google地图]]、[[Google街景]]、[[Google翻译]]和[[YouTube]]。<ref name="whitepaper2015"/><ref name="Perez">{{Cite web |url=http://techcrunch.com/2015/11/09/google-open-sources-the-machine-learning-tech-behind-google-photos-search-smart-reply-and-more/ |title=Google Open-Sources The Machine Learning Tech Behind Google Photos Search, Smart Reply And More |accessdate=2015-11-11 |date=2015-11-09 |archive-url=https://web.archive.org/web/20151110142025/http://techcrunch.com/2015/11/09/google-open-sources-the-machine-learning-tech-behind-google-photos-search-smart-reply-and-more/ |archive-date=2015-11-10 |dead-url=no }}</ref>Google指派计算机科学家，如[[杰弗里·辛顿]]和[[傑夫·迪恩]]，简化和[[代码重构|重构]]DistBelief的代码库，使其变成一个更快、更健壮的应用级别代码库，形成了TensorFlow。<ref name="Oremus">{{Cite web |url=http://www.slate.com/blogs/future_tense/2015/11/09/google_s_tensorflow_is_open_source_and_it_s_about_to_be_a_huge_huge_deal.html |title=What Is TensorFlow, and Why Is Google So Excited About It? |accessdate=2015-11-11 |date=2015-11-11 |archive-url=https://web.archive.org/web/20151110205049/http://www.slate.com/blogs/future_tense/2015/11/09/google_s_tensorflow_is_open_source_and_it_s_about_to_be_a_huge_huge_deal.html |archive-date=2015-11-10 |dead-url=no }}</ref>2009年，Hinton领导的研究小组大大减少使用DistBelief的神经网络的错误数量，通过Hinton在广义反向传播的科学突破。最值得注意的是，Hinton的突破直接使Google语音识别软件中的错误减少至少25%。<ref name="Ward-Bailey">{{Cite web |url=http://www.csmonitor.com/Technology/2015/0914/Google-chairman-We-re-making-real-progress-on-artificial-intelligence |title=Google chairman: We’re making 'real progress' on artificial intelligence |accessdate=2015-11-25 |date=2015-11-25 |archive-url=https://web.archive.org/web/20151125203839/http://www.csmonitor.com/Technology/2015/0914/Google-chairman-We-re-making-real-progress-on-artificial-intelligence |archive-date=2015-11-25 |dead-url=no }}</ref>

=== TensorFlow ===
TensorFlow是谷歌大脑的第二代[[机器学习]]系统。

从0.8.0版本（发布于2016年4月）开始原生的支持分布式运行。

从0.9.0版本（发布于2016年6月）开始支持[[iOS]]。

从0.12.0版本（发布于2016年12月）开始支持[[Windows]]系统。该移植代码主要由[[微软]]贡献。

1.0.0版本发布于2017年2月11日。虽然参考实现运行在单台设备，TensorFlow可以运行在多个[[CPU]]和[[圖形處理器|GPU]]（和可选的[[CUDA]]扩展和[[图形处理器通用计算]]的[[SYCL]]扩展）。<ref name="Metz-Nov10">{{Cite news |url=http://www.wired.com/2015/11/googles-open-source-ai-tensorflow-signals-fast-changing-hardware-world/ |title=TensorFlow, Google's Open Source AI , Points to a Fast-Changing Hardware World |last=Metz |first=Cade |date=2015-11-10 |accessdate=2015-11-11 |archive-url=https://web.archive.org/web/20151111163641/http://www.wired.com/2015/11/googles-open-source-ai-tensorflow-signals-fast-changing-hardware-world/ |archive-date=2015-11-11 |dead-url=no }}</ref>TensorFlow可用于64位[[Linux]]、[[macOS]]和[[Windows]]，以及移动计算平台，包括[[Android]]和[[iOS]]。

TensorFlow的计算使用有状态的数据流图表示。TensorFlow的名字来源于这类神经网络对多维数组执行的操作。这些多维数组被称为[[张量]]（Tensor）。2016年6月，Jeff Dean称在[[GitHub]]有1500个库使用了TensorFlow，其中只有5个来自Google。<ref name="1500repo's">[https://www.youtube.com/watch?v=Rnm83GqgqPE Machine Learning: Google I/O 2016 Minute 07:30/44:44] {{Wayback|url=https://www.youtube.com/watch?v=Rnm83GqgqPE |date=20161221095258 }} accessdate=2016-06-05</ref>

=== 张量处理单元（TPU） ===
2016年5月，Google宣布了[[张量处理单元]]（TPU），一个专为[[机器学习]]和TensorFlow[[全定制]]的[[专用集成电路]]。TPU是一个可编程的[[人工智能加速器]]，提供高[[吞吐量]]的低精度计算（如[[8位]]），面向使用或运行模型而不是训练模型。Google宣布他们已经在数据中心中运行TPU长达一年多，发现它们对机器学习提供一个[[数量级]]更优的每瓦特性能。<ref>{{Cite web |url=https://cloudplatform.googleblog.com/2016/05/Google-supercharges-machine-learning-tasks-with-custom-chip.html |title=Google supercharges machine learning tasks with TPU custom chip |accessdate=2016-05-19 |archive-url=https://web.archive.org/web/20160518235445/https://cloudplatform.googleblog.com/2016/05/Google-supercharges-machine-learning-tasks-with-custom-chip.html |archive-date=2016-05-18 |dead-url=no }}</ref>

2017年5月Google宣布第二代[[张量处理单元]]，并在[[Google计算引擎]]中可用。<ref>{{Cite news |url=https://www.blog.google/topics/google-cloud/google-cloud-offer-tpus-machine-learning/ |title=Build and train machine learning models on our new Google Cloud TPUs |date=2017-05-17 |work=Google |access-date=2017-05-18 |language=en |archive-url=https://web.archive.org/web/20170517182035/https://blog.google/topics/google-cloud/google-cloud-offer-tpus-machine-learning/ |archive-date=2017-05-17 |dead-url=no }}</ref>第二代TPU提供最高180 teraflops性能，组装成64个TPU的集群时提供最高11.5 petaflops性能。

===TensorFlow Lite===
[[2017年5月]]Google宣布从[[Android Oreo]]开始，提供一个专用于Android开发的软件栈TensorFlow Lite<ref>{{Cite web |url=https://www.theverge.com/2017/5/17/15645908/google-ai-tensorflowlite-machine-learning-announcement-io-2017 |title=Google’s new machine learning framework is going to put more AI on your phone |access-date=2018-01-14 |archive-url=https://web.archive.org/web/20170822133911/https://www.theverge.com/2017/5/17/15645908/google-ai-tensorflowlite-machine-learning-announcement-io-2017 |archive-date=2017-08-22 |dead-url=no }}</ref>。

===用于搜索引擎===
[[Google]]于2015年10月26日正式发布了{{link-en|RankBrain}}，由TensorFlow支持。

=== TensorFlow 2 ===
随着TensorFlow在研究论文上的市场份额因[[PyTorch]]的优势而衰减<ref name=":9">{{cite web|url=https://thegradient.pub/state-of-ml-frameworks-2019-pytorch-dominates-research-tensorflow-dominates-industry/|title=The State of Machine Learning Frameworks in 2019|publisher=The Gradient|first1=Horace|last1=He|date=10 October 2019|access-date=22 May 2020|archive-date=2019-10-10|archive-url=https://web.archive.org/web/20191010161542/https://thegradient.pub/state-of-ml-frameworks-2019-pytorch-dominates-research-tensorflow-dominates-industry/|dead-url=no}}</ref>，在2019年9月TensorFlow团队宣布了程序库的一个新的主要版本。TensorFlow 2.0介入了很多变更，最重要的是TensorFlow eager，它将自动微分的方案，从静态计算的图，变更为“运行时定义”的方案，它最初因{{le|Chainer}}和后来的[[PyTorch]]而流行<ref name=":9" />。其他主要的变更包括移除一些旧库，在不同版本的TensorFlow的训练模型之间的交叉兼容性，还有在GPU上性能的显著改进<ref>{{cite web|url=https://blog.tensorflow.org/2019/09/tensorflow-20-is-now-available.html|title=TensorFlow 2.0 is now available!|publisher=TensorFlow Blog|date=30 September 2019|access-date=22 May 2020|archive-date=2019-10-30|archive-url=https://web.archive.org/web/20191030134434/https://blog.tensorflow.org/2019/09/tensorflow-20-is-now-available.html|dead-url=no}}</ref>。

== 使用 ==
TensorFlow充当机器学习的核心平台和函数库。TensorFlow的API使用[[Keras]]来允许用户制作自己的机器学习模型<ref>{{Cite web|title=TensorFlow Core {{!}} Machine Learning for Beginners and Experts|url=https://www.tensorflow.org/overview|access-date=2021-11-04|website=TensorFlow|language=en|archive-date=2023-01-20|archive-url=https://web.archive.org/web/20230120082541/https://www.tensorflow.org/overview|dead-url=no}}</ref>。除了建造和训练它他们的模型之外，TensorFlow还能帮助装载数据来训练模型，并使用TensorFlow Serving来部署它<ref name=":1">{{Cite web|title=Introduction to TensorFlow|url=https://www.tensorflow.org/learn|access-date=2021-10-28|website=TensorFlow|language=en|archive-date=2023-01-20|archive-url=https://web.archive.org/web/20230120082541/https://www.tensorflow.org/learn|dead-url=no}}</ref>。

TensorFlow提供了一个Python API<ref>{{Cite web |url=https://www.tensorflow.org/api_docs/python/tf |title=Python |access-date=2022-12-12 |archive-date=2022-09-05 |archive-url=https://web.archive.org/web/20220905031826/https://www.tensorflow.org/api_docs/python/tf |dead-url=no }}</ref>，以及C++<ref>{{Cite web |url=https://www.tensorflow.org/api_docs/cc |title=C++ |access-date=2022-12-12 |archive-date=2023-01-20 |archive-url=https://web.archive.org/web/20230120082630/https://www.tensorflow.org/api_docs/cc |dead-url=no }}</ref>、Haskell<ref>[https://github.com/tensorflow/haskell Haskell]{{Wayback|url=https://github.com/tensorflow/haskell |date=20190501161337 }}</ref>、Java<ref>[https://www.tensorflow.org/api_docs/java/reference/org/tensorflow/package-summary Java]{{Wayback|url=https://www.tensorflow.org/api_docs/java/reference/org/tensorflow/package-summary |date=20170221010046 }}</ref>、Go<ref>[https://godoc.org/github.com/tensorflow/tensorflow/tensorflow/go Go]{{Wayback|url=https://godoc.org/github.com/tensorflow/tensorflow/tensorflow/go |date=20170221010654 }}</ref>和Rust<ref>[https://github.com/tensorflow/rust Rust]{{Wayback|url=https://github.com/tensorflow/rust |date=20191105112030 }}</ref>API。第三方包可用于C#<ref>[https://github.com/migueldeicaza/TensorFlowSharp C#]{{Wayback|url=https://github.com/migueldeicaza/TensorFlowSharp |date=20191105111851 }}</ref>、.NET Core<ref>[https://github.com/SciSharp/TensorFlow.NET .NET Core]{{Wayback|url=https://github.com/SciSharp/TensorFlow.NET |date=20190825220737 }}</ref>、Julia<ref>[https://github.com/malmaud/TensorFlow.jl Julia]{{Wayback|url=https://github.com/malmaud/TensorFlow.jl |date=20190512023424 }}</ref>、R<ref>[https://github.com/rstudio/tensorflow R]{{Wayback|url=https://github.com/rstudio/tensorflow |date=20190706015608 }}</ref>和Scala<ref>[https://github.com/eaplatanios/tensorflow_scala Scala]{{Wayback|url=https://github.com/eaplatanios/tensorflow_scala |date=20190218035307 }}</ref>。

== 应用 ==
广泛的应用程序使用TensorFlow作为基础，其中它已成功实现自动化图像字幕软件，例如{{le|DeepDream}}。<ref name="Byrne">{{Cite web|url=http://motherboard.vice.com/en_uk/read/google-offers-up-its-entire-machine-learning-library-as-open-source|title=Google Offers Up Its Entire Machine Learning Library as Open-Source Software|accessdate=2015-11-11|date=2015-11-11|archive-url=https://web.archive.org/web/20151123090522/http://motherboard.vice.com/en_uk/read/google-offers-up-its-entire-machine-learning-library-as-open-source|archive-date=2015-11-23|dead-url=no}}</ref>2015年10月26日，Google正式启用了由TensorFlow提供支持的{{le|RankBrain}}。RankBrain现在处理大量的搜索查询，替换和补充传统的静态算法搜索结果。<ref name="Woollaston">{{Cite web|url=http://www.dailymail.co.uk/sciencetech/article-3311650/Google-releases-TensorFlow-Search-giant-makes-artificial-intelligence-software-available-public.html|title=Google releases TensorFlow – Search giant makes its artificial intelligence software available to the public|accessdate=2015-11-25|date=2015-11-25|archive-url=https://web.archive.org/web/20151125213833/http://www.dailymail.co.uk/sciencetech/article-3311650/Google-releases-TensorFlow-Search-giant-makes-artificial-intelligence-software-available-public.html|archive-date=2015-11-25|dead-url=no}}</ref>

==参见==
* [[人工神经网络]]
* {{tsl|en|Comparison of deep learning frameworks|深度学习框架比较}}
* {{le|深度学习软件比较|Comparison of deep learning software}}
* [[深度学习]]
* [[机器学习]]
* [[卷积神经网络]]

== 引用 ==
{{Reflist|30em}}

== 外部链接 ==
* {{Official website|https://www.tensorflow.org/}}
* {{Official website|https://tensorflow.google.cn/}}{{zh-cn}}
* {{github|tensorflow/tensorflow|TensorFlow}}

{{深度学习软件}}
{{Differentiable computing}}
{{Google AI}}

[[Category:深度学习软件]]
[[Category:自由統計軟件]]
[[Category:开源人工智能]]
[[Category:Google軟體]]
[[Category:用C++編程的自由軟體]]
[[Category:用Python編程的自由軟體]]
[[Category:使用Apache许可证的软件]]
[[Category:Python科学库]]
[[Category:2015年软件]]

TensorFlow - 版本历史

imported>Ohtashinichiro：​/* 外部链接 */

imported>Ohtashinichiro：/* 外部链接 */