TensorFlow What can be done ? Give Way Google Brain Chief engineer tells you
edit | Mingming
1 month 19 day , At the geek Park innovators Conference IF2018 The scene of ,Google Brain Chief engineer Chen Zhifeng published a report entitled ：《 The answer starts with defining the question ——TensorFlow
What can be used for ?》 Speech , Shared Google Brain Research direction in recent one to two years , And in the TensorFlow Some work on , achievements , progress . The following is a transcript of the speech ：
Deep learning is very popular these years , stay Google In the search traffic , Deep learning in the past 7-8 Years , It's increased by about 100 times , From this aspect, it also reflects the academic and industrial attention to this technology , It's improving rapidly .
Everybody wants to know , What is deep learning ? What can it do for us ? How to apply this technology to the actual scene ?
Deep learning is not a sudden technological field , Its core algorithm is neural network . Neural network is a machine learning model , The main characteristics of this model are , It can fit any mathematical function , In particular, a similar method of iterating big data is used to train the model .
With the popularity of big data , The development of computer hardware computing power , There are also breakthroughs in the algorithm itself , You suddenly found out , This deep learning technology based on neural network model , In many application fields, it can greatly improve the tasks that the software system needs to complete in the past .
You must be familiar with the most classic application fields , It's image recognition . since 2012 Since , In this area , Error rate of automatic image recognition 20% All the way down to 4% about , It has exceeded the ability of an ordinary person to distinguish images .
Google Deep learning technology has been applied to many products many years ago , In the process , We also iterate , Several generations of software systems supporting deep learning have been developed , Eventually, it led us to 2015 year 10 Open source in January TensorFlow, Hope to further promote the application and research of deep learning .
TensorFlow Now it has evolved into a fairly complete open platform for deep learning software .
Platform supporting multiple hardware
for instance , It supports CPU,GPU Hybrid training platform for data center , It also supports models that train data centers , It is relatively convenient to deploy to different mobile applications , Can support similar Google Self developed TPU processor .
This multi platform support , It can help the most users and application scenarios , At the same time, we are very grateful to many colleagues in the industry for their support , In the United States, for example Intel And NVIDIA are helping us optimize TensorFlow Performance on their respective hardware .
Support multiple development environments
A platform that supports a variety of hardware is the foundation ,TensorFlow It's always been a goal , Is to help as many developers as possible , Be able to use the technology of deep learning , Finally, it makes the majority of users benefit from the ability , Based on this idea ,TensorFlow We always attach great importance to the support of various programmer development environments . for instance , Developers can use it in the main development environment TensorFlow.
TensorFlow In our company's internal application promotion is very comprehensive and thorough , for instance TensorFlow Help early Google Core business （ Search and advertising ）, The model of deep learning is implemented , And in the core business reflects their performance .
In spam filtering , We used it, too TensorFlow Training model , At the same time, the android app recommendation , It's online, too TensorFlow And so on . quite a lot TensorFlow Applications are all in the background , Most users may not have a direct experience .
Go deep into mobile terminal
Let me give a few examples on the mobile terminal , In fact, deep learning has directly affected thousands of users .
Android mobile phone self timer function
for instance , The latest version of Android phones has added a self timer feature , This self timer function is a visual model trained by applying deep learning , The foreground and background pixels can be separated very accurately , The foreground pixel and background pixel are processed respectively , In this way, the background can be virtualized .
Realize this function , Traditionally, mobile phone manufacturers need to add a second camera , This will increase the cost of mobile phones , At the same time, existing mobile phones are not easy to achieve this effect , Through a new algorithm , We can achieve something that could have been very expensive in the past .
Same as image processing , Speech processing is another area that has been profoundly changed by deep learning , Neural network is used earlier than image processing in speech recognition .
Products , In recent years, the main voice of smart speaker is popular , One of the reasons is that deep learning algorithm greatly reduces the technical threshold of speech recognition and speech generation , It may have been necessary in the past 20—50 Only a team of doctors can complete the task , Now download an existing model to customize it , The same effect can be achieved . General machine learning framework , Can help more developers , Develop voice applications suitable for specific application scenarios .
Another example , Now you can take a picture with your mobile phone , Mobile phone software can automatically recognize the text in the graphics , Translating words into another language , Seemingly simple application , In fact, it is a natural combination of image technology and machine translation technology .
More than a year ago , I had the privilege of being part of Google , Upgrading the past translation system to a neural network based system , That upgrade greatly reduced the error rate of machine translation . Translation between some of these languages , It can almost achieve the effect of artificial translation
Email auto reply function
Using deep learning technology , We can not only greatly improve the function and performance of existing products , We've also developed new features that were hard to imagine in the past , For example, the function of automatic email reply , Email software on Android can analyze users' email .
For example, you received an email from your friend this morning ,“ Where are you going to eat in the evening ”, In most cases, there are only three possible answers ： I'll be there on time ; I'm sorry I can't come ; I'm free , But I may need to be late .
Our email will automatically provide you with three choices , In this case, I see the mailbox on the road , You can reply to an email with one click , It greatly facilitates the efficiency of email processing on the mobile phone .
TensorFlow Applications in other fields
Through the previous many application examples , We can see that , A general deep learning framework , It can help many existing Internet applications to improve the level of intelligence , But we can also see that ,TensorFlow Such a deep learning framework , It can also be applied in many other fields .
Like our colleagues in London , Efforts have been made in the past two years AlphaGo The work of , This is because TensorFlow The help of the framework itself , because TensorFlow Supercomputer clusters can be used , Support the latest accelerators ,AlphaGo The team can focus more on algorithms .
Intelligent medical treatment
There are also people in our group who are doing intelligent medical work . because TensorFlow It is a general framework , They can easily reuse existing image recognition models or natural language processing models , For specific application areas and data , Retrain and fine tune the model , You can get more than 95% The accuracy of , This accuracy rate has surpassed that of ordinary ophthalmologists 91% The level of .
In terms of autonomous driving , We are here Alpha Waymo My colleagues , It's also being used TensorFlow Continuous improvement of depth models in automated driving systems , Including the segmentation of road scene , Radar signal processing and so on .
Baby food making
We are particularly pleased that , By putting TensorFlow Open Source , The threshold for us to use deep learning technology has been greatly reduced . In the past two years, we have seen a lot of industries that have nothing to do with the Internet , Also began to try to use the technology and methods of deep learning , For example, this company is a baby food company , Introduced TensorFlow Trained intelligent system , This intelligent system can classify the ingredients of baby food , Remove some rotten apples and bananas more accurately , This can accurately control the quality of baby food .
scientific research —— Space
We are also excited to see ,TensorFlow It is also used in scientific research other than computer science , Last month, for example, NASA announced a study , NASA has a Kepler program , Their scientists and our colleagues have developed one TensorFlow Model of .
Kepler project itself aims to continuously observe the brightness changes of stars in space through telescopes , Discovery of planetary systems outside the solar system , Finally, we hope to find another planet suitable for human habitation . At present, the program has accumulated more than 10 billion observational data , A few months ago this TensorFlow Model of , Help scientists find out 2500 Kepler beyond light years 90 The eighth planet in the galaxy .
scientific research —— ocean
People don't just look up at the stars , At the same time, we are looking back at our earth , In the example of this application , Scientists in Australia TensorFlow Developed image recognition model , Among the tens of thousands of aerial photographs of the ocean , Fast and accurate identification of large marine mammals in need of protection , For example, the rare animals manatee .
scientific research —— jungle
Scientists also use it TensorFlow Speech processing technology is applied to bird protection . They installed a lot of microphones in the jungle , Collect the sound of birds , The model can accurately estimate the number of birds in a forest , So as to protect them more precisely .
Our group and colleagues have made some very interesting applications , They are trying to use deep learning technology to create music , It's interesting , The music created by these music , Also received a professional DJ The affirmation of .
Increase support for Chinese market
since 2015 Since the open source , We've been trying to increase the TensorFlow Investment , Recently, we began to increase our support for China , For example, we're building one TensorFlow Chinese website of ,TensorFlow We've also seen rapid growth in developers , meanwhile , We also see more than 1000 Many people participated TensorFlow Development of , It's a very active community .
in addition , Our data also show that , So far, the world 180 Multiple countries , Various users have downloaded more than 1000 Ten thousand times TensorFlow Development package , This shows that TensorFlow Application development is also very common .
The picture above also tells you , There are a lot of developers in China who are paying close attention to it TensorFlow, in fact , We are building partnerships with many Chinese companies , Actively support and help them use it better TensorFlow.
For example, Jingdong has built an internal structure TensorFlow Training platform , For developing images , Natural language related models , And use them in customer service advertising and other fields . Xiaomi is also trying a similar technical route , Support their special applications on the ecosystem . Netease's Youdao notes , Netease translation Jun also used TensorFlow Models of vision and language .
Thank you very much to our users and partners TensorFlow Feedback ,TensorFlow Efforts are also being made to develop new features , We developed one last year TensorFlow pattern , This model can be more conducive to the front-end development , Easy to debug , Support more dynamic programming modes .
There's a very short program here , In this program , You should be able to see that the most important feature of this pattern is in the front-end program flow , It can reflect the process of program algorithm logic more directly , We believe that this will be particularly helpful for Rapid Prototyping Development and debugging .
in addition ,TensorFlow The main projects in the past year have been pushed forward TensorFlow Lite pattern , This model is specifically aimed at migration 动和嵌入式应用场景打造的机器学习平台,它的目标是希望把在云端训练的机器学习模型,更加简单,高效的迁移到移动端上进行部署.
干货 | AI 工程师必读,从实践的角度解析一名合格的AI工程师是怎样炼成的
详解 | 如何用Python实现机器学习算法
经验 | 如何高效学Python?