November 15th,Google releases mobile device version of machine learning framework, TensorFlow Lite PreviewThis means that the trend of putting artificial intelligence into our mobile phones is a step forward.
Google in May this year on the I/O Google, has announced the TensorFlow Lite this tool. At present, TensorFlow Lite supports a lot of training and optimization models for mobile terminals, including visual model, picture recognition model and device dialogue model. Both Android and iOS platforms can be used by developers.
Bloomberg reports last month,Google wants to promote its own open source artificial intelligence framework TensorFlow in ChinaTo return to the Chinese market by reducing the difficulty of developing and using artificial intelligence.
Maybe non developers are not familiar with TensorFlow, but you should have heard of the name DeepMind, which has developed AlphaGo, which beat this year, and AlphaGo Zero, who can't find an opponent in the go world for 3 days.
DeepMind announced in May 2016 that they would replace their open source machine learning platforms and switch from Torch to TensorFlow. It can be said that the two AlphaGo and AlphaGo Zero, as well as more artificial intelligence released by Google, finally in the TensorFlow "nurturance".
Google announced and announced open source artificial intelligence system TensorFlow at the end of 2015. China is one of the fastest growing TensorFlow users in Asia, even if it fails to use a large amount of Google services in china. In April, TensorFlow was in chargeRajat Monga expresses to mediaTensorFlow has downloaded more than 140 thousand downloads in china.
stayTensorFlow Chinese communityHome page, there is a sentence:
TensorFlow is an open source artifact for artificial intelligence.
On GitHub, TensorFlow has become a popular machine learning open source project since last year.GitHub's 2017 developer ReportTensorFlow is the most fork project, and 7300 of them have contributed to TensorFlow.
(TensorFlow is the most fork project)
The number of developers participating in TensorFlow ranks fifth
TensorFlow in China
Since October, Google developer community developer Festival has been held in many places in ChinaGDG DevFestThis is an activity initiated by Google and organized by GDG around the world.
Love fan children (WeChat: iFanr) in the Google community of Guangzhou developers Festival, which in the Google work for many years and has been in the Google Brain developer Liao Baohua, to the presence of hundreds of developers, share of non professionals, machine learning beginners how to make more efficient use of TensorFlow to develop.
Liao Baohua said, TensorFlow allows developers and enterprises to stand on the shoulders of giants and build their own models efficiently. He gave an example of building machine learning models, compared to building houses.
We don't want developers to build bricks and mortar like bricks and mortar, but to assemble several modules that are already in place.
Site developers said that they are also using the TensorFlow interface (API) to train the model, including many suppliers, live broadcast, unmanned aerial vehicles and other industries developers.
GDG, the organizer of Nicky in Guangzhou, told, iFanr, that TensorFlow greatly reduced the threshold for developers to use machine learning, encapsulating a large amount of API, so that developers did not need to do a lot of data simulation. And that is, a lot of TensorFlow interface, in the country can be used normally, this and Google, Play and other use is very different (you know).
(screenshot from TensorFlow official website)
Do not think that TensorFlow is too far away from us, there are many domestic large electricity supplier in the use of customer service is based on TensorFlow development. In TensorFlow's official website, we can see that the domestic millet, ZTE, Jingdong and other enterprises, but also the use of TensorFlow.
And Google Brain released itArtificial intelligence that can help you paint"Quick, Draw!" and "AutoDraw" are also developed based on TensorFlow.
(cat painted with AutoDraw)
With TensorFlow, you need a lot of hardware
Although TensorFlow is a popular framework for deep learning, there are many domestic developers who believe that if you need to select an artificial intelligence framework, use your own data to train a deep learning model,TensorFlow is not necessarily their first choice.
In addition to a large number of Google products and services in the country can not be used normally, there are such artificial intelligence, deep learning open source database, there are many alternatives, such asCaffeKaras and so on.
I'm taking a more active part in the affairs of china.
That said, but Google's return to China should still be an illusion. Because TensorFlow is not indispensable to ordinary developers. And for the senior enterprise users, especially Internet companies, most of the early years began to set up research and development of artificial intelligence, and other enterprises, will also give priority to the use of Baidu, Ali cloud these domestic services.
In November 15th, the Ministry of science and technology announced the first batch of national artificial intelligence open innovation platform list:
Relying on the construction of Baidu Inc autopilot open innovation platform of national city relying on artificial intelligence, artificial intelligence brain state open innovation platform construction aliyun company, relying on the construction of open innovation platform for medical imaging state of artificial intelligence based on Tencent Inc, iFLYTEK company building intelligent voice state artificial intelligence open innovation platform.
However, with TensorFlow, you need to use enough hardware to do. Google (Tensor Processing Units) launched by TPU is a hardware accelerator specially designed for TensorFlow. At present, the second generation of TPU (also known as Cloud TPU) has the ability to train machine learning models and deal with reasoning tasks two capabilities.
However, some people think that although the Internet giant to create ecological and not the chips have been rare, but the launch of Google TPU and not to be with NVIDIA, NVIDIA, Cambrian these manufacturers to compete, but in order to highlight the power of TensorFlow.
October, GoogleThe Pixel2 mobile phone has an image processing coprocessorThat is, ImageProcessing Unit (IPU). The IPU is primarily used to speed up machine vision and machine learning, and it supports accelerated programming languages, including TensorFlow, of course.
Life writes good stories work mailbox: email@example.com