Shulou(Shulou.com)06/02 Report--

For more than a hundred years, information technology has become an infrastructure from its infancy, which has made countless brilliant science and technology giants. Today, artificial intelligence is also on the road to social universal technology.

It is not easy to predict the navigation chart of AI from the industry law of the last wave of information technology. But when we go back to the scene of history, we can also find out some truth.

For example, it is a widely accepted fact that almost all famous technology companies have laboratory support, large or small, behind them. Many of them have made great contributions to the field of human communications, and there is no lack of Nobel Prize and Turing Prize winners.

It is no exaggeration to say that top laboratories exist, such as those of Shockley, the first citizen of Silicon Valley, Bell Labs of AT&T, IBMResearch Labs of IBM, PARC of Xerox's Palo Alto Research Center. And so on, has established the innovation water level of Silicon Valley, but also provided the fertile technical soil for the follow-up nearly hundred years of information industry revolution.

And this kind of industry development inertia, also naturally extended to the field of artificial intelligence. Google artificial Intelligence Lab, which gave birth to the machine learning framework TensorFlow, Microsoft Research, which incubated intelligent assistants Xiao Bing and Xiao Na, or various laboratories and research centers of many domestic giants, have become an important support for the intelligent economy.

Among them, YouTu laboratory, as one of the twin engines of Tencent AI technology, is an existence that can not be ignored. Different from other laboratories, YouTu, who has undergone three important iterative upgrades since its establishment in 2012, can be said to have completed the century history of the laboratory in the communication era in eight years.

When we try to further solve the basic problems such as the historical opportunity points of science and technology enterprises and how to bridge the gap between basic research and the commercial value of products at the high speed of AI, YouTu's road of upgrading and training is an excellent reference frame.

The rising Sun: the technological "Breakthrough" in the quiet Corner

The roles of laboratories are also different in different industrial stages.

For example, in the early stage of the development of the industry in the ascendant of communication technology, relying on the laboratory to complete the bottom technological breakthrough and lift the ceiling of enterprise competitive advantage is the most common demand and the original intention of the establishment of the laboratory.

So we can see that in recent years, tech giants have been recruiting AI bigwigs to join the industry. Google Labs has hired the most important expert-Yann LeCun, the artificial intelligence pioneer and head of the artificial intelligence laboratory of Turing Award winner Geoffrey Hinton;Facebook, is also a legend in the AI world.

Attracting the yearning and following of scientific and technological workers not only costs money (high annual salary), but also can not see the benefits quickly. For commercial organizations that put corporate interests first, why do they do this?

To find the answer, we might as well go back to 1925 and explore the core value of technology research to technology companies from the most successful Bell Labs.

As many people know, AT&T spent 3% of its output value on Bell Labs' research and development, and Bell Labs, which had no shortage of funds, became the first choice for scientists in the information field at that time. Of course, their joining has also contributed inestimable value to AT&T, such as the invention of information theory, the launch of the first communication satellite, the establishment of the first commercial optical fiber, and so on. Numerous innovations have made AT&T lead the world in the field of communications for a long time.

The importance of technology research to the first-mover advantage of business value can be seen in Tencent's AI layout.

From the initial layout to the full application of AI in its own business and cloud services, Tencent has something to do with the research of YouTu Lab.

At the beginning of its establishment in 2012, YouTu Laboratory first did the secondary compression of pictures, and soon shifted its research focus to face recognition and detection, and now focuses on the whole computer vision to carry out related technology research and development and reserve.

In the 1.0 stage, the core of YouTu is to raise the technical level of Tencent and constantly improve on the visual AI.

In 2015, YouTu developed YouTu Uface face recognition algorithm based on the concept of deep learning, and evaluated it on LFW database in the same year, ranking first in the world with 99.56% accuracy.

In 2017, YouTu developed the "YouTu grandmother model", with a depth of 10 to 1000 layers, which can deal with all scene data and solve all kinds of "heterogeneous data fusion".

These technological achievements have set world records many times, and Tencent, which has won a lot of AI patents, has also been promoted to the first square of global AI technology research and development.

Perceiving the profound changes in technology and capturing industrial changes through active exploration is the key significance of YouTu's connection with the AI world in stage 1.0, and it also takes the lead for Tencent to enter the intelligent economy.

The scorching sun is like fire: the supporter of mining rich commercial mines

The change occurred in 2018.

In September 2018, YouTu Lab was upgraded to Tencent computer Vision Research and Development Center, thus entering the 2.0 stage.

On the one hand, Tencent further increased its R & D investment in YouTu; at the same time, YouTu anchored the two-track route of "basic research + industrial landing".

Tang Daosheng, then senior executive vice president of Tencent, mentioned that YouTu focused on exploring the "four major directions" of social entertainment, industrial production, social progress and cutting-edge exploration. in-depth scene innovation around the "ten major areas" such as medical care, self-driving, industry, retail, office, culture, and social welfare.

Is it reliable to "run high marks" in technology and "run scenes" in business?

Back in the 1990s, communication technology from the exploration period to the practical and mature period, a major sign is the commercialization of the laboratory.

IBM Research, for example, was only 1/10 of the size of Bell Labs in its heyday, but it was so innovative that it became the largest patent applicant in the world at that time. The characteristic is the combination of basic research and industrial application.

At that time, the IBM lab cut down some research that was more theoretical than cost-effective. Once a research project is available, move the entire research team from the lab to the product department.

It has to be said that today, when AI is so close to the commercial market, the progress of technology and the integration of business scenarios are more inseparable.

In 2018, the environment faced by YouTu's laboratory is that artificial intelligence technology has completed the primitive accumulation of algorithms, and revolutionary breakthroughs began to decrease.

At the same time, various industries begin to realize the value of artificial intelligence, and the industrial side has a strong desire for technology blessing. During this period, how can the laboratory complete the systematic road from technological basic development to industrial deployment exploration?

After the upgrade, YouTu chose the hug scene.

YouTu's engineers said that at that time, YouTu felt that YouTu was not an independent laboratory, but more like a Tencent internal business support department. According to the needs of various business lines and B-end enterprises, to do the corresponding technical support.

Tempered by the rich scenes in Tencent, with the advantage of gathering data resources, the integration of YouTu's basic technology and scene is completed quickly. Many basic researches related to face recognition and image recognition are packaged as SDK or standard services and delivered through industry solutions.

In internal business, it has landed in more than 90 self-owned product lines, such as QQ, QQzone, Wechat, Weishi, Tencent Cloud, etc.

At the same time, the industry benchmarking cases of cooperation with corporate customers also flourished at this stage.

For example, together with Ningbo Bank to promote the application of AI technology in the financial field; with corporate users to launch intelligent underwriting and intelligent solutions with OCR as the core; WeChat Pay's facial scanning payment function was put into use in Shanghai Carrefour; assisted Huaxing Optoelectronics more than 100 industrial quality inspection, power grid intelligent inspection in cooperation with China Southern Power Grid. Wait.

In the words of YouTu employees, "customers sometimes don't know what their business wants, so YouTu and other architects and project managers join in to analyze how to use visual AI in the project."

It can be said that in the ferocious wave of AI actively integrating into the industrial perspective, the two-legged model of "basic research + application landing" put forward by YouTu laboratory has become an important guarantee for Tencent AI to run through the new industrial era.

AI Gao Zhao: ubiquitous intelligent lifter

At the recent World Conference on artificial Intelligence, new changes have taken place in the "human setting" of YouTu's laboratory.

On June 10th, YouTu Lab announced that it officially opened up TNN, a new generation of mobile deep learning reasoning framework. This news reveals two propositions:

1. Tencent YouTu Lab further open source, the role of AI ecological promoter continues to be solid.

two。 The threshold for mobile AI development has been further lowered, and Tencent has adopted a clear posture of a new generation of developers.

What is the opportunity for upgrading this time?

It is not so easy to find a standard laboratory in the field of computer communication. Because the most essential difference in the logic between the communications industry and the AI industry is that artificial intelligence is open source from the very beginning.

AI technology determines that it needs an open ecological environment to accelerate the growth of technology; in addition, innovation and iteration from real applications is an indispensable driving force for the prosperity of AI.

Therefore, open source has become the consensus of "further achievement feedback" in many laboratories around the world. AI giants such as Google, Apple, Facebook, etc., all have an active layout in the open source community.

And YouTu of upgrade 2.0 has been in the global AI open source ecological action constantly.

For example, in strategic cooperation with Science, the official publication of the American Association for the Advancement of Science, we share resources and information in the field of computer vision, and directly open our formed technology platforms, models or neural networks to global scientists on the platform, so as to jointly promote the development of CV technology.

For example, with the help of Tencent YouTu Lab open platform, face recognition, image recognition and other world-leading algorithm capabilities, open calls for users through cloud services.

In addition, the Rapidnet forward computing framework, the RapidAIoT edge computing framework, and the open source of ncnn, the industry's first mobile-focused inference framework, make it easier for more developers to touch AI.

So 3. 0 stage, YouTu's open source is so different?

On the one hand, YouTu occupies the "bridgehead" zone at the level of basic research and assumes more platform responsibilities in industrial coordination.

In 2020 alone, YouTu was accepted 17 papers by the CVPR of the top international CV conference. These leading innovations, through open source and platform products, release greater AI value, which is the social significance that YouTu is eager to undertake in stage 3.0.

Therefore, we see that the formal open source new generation mobile deep learning reasoning framework TNN, with its excellent performance, lightweight development, simple and practical features, makes the mobile AI open to more developers.

On the other hand, the development of mobile Internet tells us that infrastructure platform is a strategic competition. Take Android and IOS ecology as an example, if you want to further activate the value of business innovation, you need to attract the vast majority of AI developers and users to their own open source environment.

Only the more developers who use their own tool chains and tool ecology, the more likely they are to be caught up in the expansion of the business ecology.

From this perspective, YouTu's first open source deep learning framework TNN makes it easier for developers to develop artificial intelligence App, which is of great significance to nourish the end-to-side business ecology and promote the innovative application of AI in end-to-side scenarios.

This is worthy of greater expectation for making AI a ubiquitous necessity in the intelligent age, as well as water, electricity and WiFi as the background of modern life.

Since its establishment in 2012, YouTu's eight years have also been the rise of this wave of artificial intelligence from its inception, outbreak and rational integration.

YouTu from the "technology-platform-scene" upgrade practice, so that the visual AI to practical exploration, take root in the industrial field. The whole journey is like the epitome of the butterfly change of information technology for a hundred years.

Laboratory model, zenith technology, business society, also at this moment, let YouTu become the most intuitive shadow of the intelligent future.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.