Shanghai Lingang, a new artificial intelligence computing center (AIDC) representing the strongest computing power in Asia, will be completed by the end of the year. If all its computing power is used for video processing, the video duration that this facility can process in one day can reach 23600 years.
From today to 23600 years ago, mankind is still in the Paleolithic Age.
Of course, powerful computing power is not simply used to process video. In the future planning, this computing power center can simultaneously meet the needs of four smart cities with a population of more than 20 million.
But this is only part of a larger plan. In July this year, Xu Li, co-founder and CEO of Shangtang Technology, formally introduced the new AI infrastructure built by Shangtang Vision – SenseCore Shangtang AI device at the World AI Conference. The AIDC near the port and Shangtang’s existing supercomputing center together constitute the computing power foundation of this “big device”.
The upper-middle layer of Shangtang’s “AI big device” is an algorithm platform that can realize AI mass production and deployment applications, and an algorithm library for enterprises and developers. Through this “big device”, the training and deployment of AI algorithm is no longer a “craft”, and the innovation of AI will fully realize cost reduction and efficiency increase.
For AI technology, landing in different industries and fields no longer requires a lot of manpower. For Shangtang Technology, the emergence of AI devices means that the company is one step closer to the vision of setting goals when starting a business.
Established 7 years ago, Shangtang has invested in polishing the underlying technology
Back in 2015, the algorithm developed by Shangtang team won the championship in the large-scale visual recognition challenge of ImageNet for the first time. The accuracy rate of the algorithm set a new record, 6 points higher than Google’s performance the previous year. The algorithm was published on CVPR, opening the curtain of AI promoting industrial revolution. This is an important milestone in the development of Shangtang’s original technology.
Insisting on originality has always been the label of Shangtang Technology. Starting from the bottom, there are many advantages in making original technology. Xu Li believes that “in the development of enterprises, there will be more opportunities to do the bottom technology. The breakthrough in technology will create the blue ocean market; the leading in technology will bring the different quality of products.”
Algorithm, data and computing power are the three underlying elements of deep learning technology. In order to build the underlying core advantages of deep learning and original technology, Shangtang Technology has quickly devoted itself to two things after its establishment: recruiting talents and building its own supercomputing center.
In the first two years after its establishment, the top Chinese scientists in the field of deep learning almost concentrated in the office of Shangtang Technology, with a number of more than 150 people. Their goal is to move the academic algorithm innovation of deep learning to industry, and constantly consolidate the original deep learning framework SenseParrots. Compared with the popular open source framework in the industry at that time, SenseParrots has a higher utilization rate of computing power and can efficiently train AI algorithms with stronger performance and more types, which has also become the cornerstone of innovation of Shangtang algorithm and the essence of today’s AI large device platform layer.
The self-built supercomputing center is also the long-term layout of Shangtang Technology, in order to have its own computing power scale. It is understood that the founding team of Shangtang has built a special supercomputing center for AI training from the beginning. This is also the predecessor of AI large device computing power.
With these two foundations, Shang Tang won the championship of ImageNet Computer Vision International Challenge in 2015, and even developed 1207 layers of the world’s deepest deep neural network to win the championship again in ImageNet in 2016. This can be seen as the pioneer exploration of Shang Tang’s later exploration of AI big model. At that time, Xu Licheng said, “Shangtang has been able to continuously challenge the industrial red line and create new application scenarios because of the three ‘most’ – the first Chinese team to engage in in-depth learning research, the largest in-depth learning research team in Asia, and the enterprise with the highest density of doctors.”
In the early stage of AI entering the industry, that is, from 2015 to 2016, the industry saw more market potential of AI technology in solving space safety, traffic management and other aspects. However, with the persistence and understanding of original technology, Shangtang has a longer vision. Since its establishment, they have taken a grand goal of “let AI lead human progress” as their mission. “Let AI empower all industries” is an important way to achieve this mission.
Because of this, Shang Tang hopes to be an “all-around player” of in-depth learning, so that he can transmit AI capabilities to wherever he needs AI like a power plant. The positioning of Shangtang Technology AI Platform Company is also born from this.
Xu Li also summarized Shangtang’s business vision as two sets of tools. He once said, “The pain point of traditional enterprises is efficiency and innovation in the final analysis. Therefore, Shangtang Technology is making two tools, one is productivity tool to reduce costs and increase efficiency for traditional industries; the other is interactive tool to bring new interactive experience and form new application scenarios.”
However, whether it is “enabling all industries” or “two tools”, it is easy to understand in the literal sense, but it is not easy to put it into industrial practice successfully. This requires AI enterprises to have a closed loop from core technology breakthrough to rapid commercialization.
Xu Li pointed out that the closed loop of Shangtang consists of five links: first, find the most real and effective scale rigid needs of the industry and define the problem; Secondly, we should make breakthroughs in basic technology based on the technical red line required by the industry; Third, cooperate with effective data closed-loop to form positive feedback on technology, data and business; Fourth, transfer core value elements of traditional industries with a more advantageous business model, promote industrial upgrading, and drive rapid and large-scale application landing; Finally, the economic value brought by large-scale commercial application provides a solid foundation for the continuous technological innovation capability.
Adhering to this business path from technological breakthrough to industrial landing, Shangtang has successfully empowered its original technology to all walks of life. At the same time, with the continuous improvement of the underlying development tool chain, Shangtang has gradually acquired the general algorithm production capacity of the pipeline. Since two years ago, “algorithm factory” has become a new symbol of Shangtang. Today, Shangtang has developed more than 21000 algorithm models, thus forming the algorithm layer of the AI device.
So far this year, Shangtang’s AI technology has widely covered multiple industries and scenarios in the four sectors of smart city, smart business, smart life, and smart car, and is in a leading position in multiple vertical fields.
Xu Li believes that when the speed of innovation is fast enough to truly combine innovation with industry, there will be a speed that exceeds the average speed of industry innovation. This speed is the “escape speed”.
Facing the future
Shang Tang gets through the combination of virtuality and reality and creates a new paradigm
The birth of SenseCore Shangtang AI large device is undoubtedly the best embodiment of Shangtang’s profound accumulation.
It represents Shangtang’s adherence to original technology in the past seven years, and is the essence of the combination of underlying computing power, complete algorithm production tool chain, and massive algorithm models for different industries.
It opens a new “AI industrial era”. It will become the infrastructure for AI development in the next decade or even decades. With its strong generalization ability, it will enable thousands of industries to produce the AI capabilities required for lower cost and more efficient production, and enable AI technology to bypass by analogy, connect points and scale rapidly.
The Chinese Academy of Information and Communication pointed out in the White Paper on the Core Technology Industry of Artificial Intelligence that the current intelligent technology is infiltrating into more industries, the process of industrial scale development is accelerating, and the AI industry has reached the critical point before the outbreak of scale economy.
How big is this scale? According to McKinsey’s prediction, by 2030, about 70% of industrial enterprises will use artificial intelligence technology, which is expected to add 13 trillion dollars of added value to the world.
Niu Kui-guang, partner of IDG Capital, once said: “The landing of AI technology has entered the deep water zone, and how to build a common paradigm to break the application limitation of single-point scenarios has become the test question that all AI practitioners must face at present. Tackling difficulties needs to gather the wisdom of the group. Since its establishment, Shangtang has continuously optimized its own capabilities from multiple dimensions such as software layer and hardware layer by virtue of its deep talent reserves. I believe that in the future, Shangtang will continue to improve AI with its own innovation ability and solid technical foundation Technology empowers all industries and makes each industry rejuvenate in the digital economy. ”
In the era of large-scale AI technology, AI will fully penetrate into all chains and links of the industry. The digital world composed of information will also achieve seamless connection and interconnection with the real world. At this time, Shangtang’s productivity tools and interaction tools will no longer be independent individuals, but will be combined into a new tool of “virtual and real integration”. The former will enable traditional industries to “go from real to virtual” and fully realize the digitalization of the physical world, while the latter will develop a new “go from virtual to real”, let virtual shine into reality, form a new innovative paradigm of “virtual and real integration”, and realize the value closed-loop of the digital economy.
At this year’s World Conference on Artificial Intelligence, Xu Li summarized the realization of this paradigm as “three steps”. The first step is the digitization of scenarios – this is the stage the industry is going through. That is, through various sensors and other devices, the information of the physical world will be pixelated and 3D, and superimposed on the digital twin space;
The second step is the structuring of elements, that is, through the perception and understanding of AI, people, things, things and fields are structured and extracted to become meaningful elements of business;
The third step is the interactivity of the process, that is, by accessing, modifying and interacting with the elements in the digital twin space, feedback and influence the real business process, and realize the process reengineering.
With the support of “AI big device”, this set of innovation paradigm has been fully penetrated into the four sectors of smart city, smart business, smart life and smart car.
Shangtang Technology participated in the construction project of “One Network Management” in Shanghai, which clearly demonstrated the innovative achievements of “from reality to reality”.
You can understand it as a “digital twin Shanghai” that projects reality into the virtual world. It keeps pace with the real world, and at the same time allows city managers to break through the constraints of physical space and make more accurate decisions and deployment.
Take the deployment of Jiangsu Road Street, Changning District, Shanghai as an example. With the transformation of AI scene segmentation and other technologies, 1000 cameras in the street have become intelligent perception neurons, always paying attention to various problems in the area. Problems such as garbage exposure, illegal operation on the road, water accumulation on the road and illegal parking of vehicles can be quickly identified from the image, and then the order will be immediately sent to the front-line staff for treatment. The work mode of the street grid supervisor has also been transformed from the previous network patrol mode to the AI intelligent search, recommendation and order distribution precise governance mode.
The above scenario contains a large number of low-frequency long-tail requirements. Xu Li pointed out that nearly 80% of big data applications have only completed the first step of data, and have not combined data with real business processes. Although meeting these needs has great commercial value, it faces huge cost pressure in terms of human input, data collection and computing power. The value of “AI big device” is to reduce the price of artificial intelligence production factors in a real sense, and significantly reduce the difficulty and threshold of factor structuring and process interactivity.
In addition to the smart city, the model of intelligent search and recommendation by AI is also creating new imaginative space in the fields of community, commerce, medical care and so on.
Taking the Tianlin Street Community in Shanghai as an example, the application of AI technology in community management has realized the remodeling and optimization of various processes. For example, it can identify the potential safety hazards that residents have paid attention to in recent years, such as the electric car entering the elevator, open fire, and occupying the life passage, and automatically make corresponding measures to ensure the efficient, safe and orderly operation of the community.
Schindler, the world’s leading escalator enterprise, can structure all kinds of scenarios on the elevator by integrating the AI technology advantages of Shangtang in the fields of object detection, behavior recognition, density analysis, etc., define events such as passenger retrograde, pushing strollers, wheelchairs, carrying large luggage, or passengers falling down, and timely alarm when the event occurs, and take actions such as reducing the running speed or stopping the elevator. The entire interaction process can be completed automatically, reshaping the past passive response mode.
The medical industry also benefits from this. The SenseCare intelligent diagnosis and treatment platform originally created by Shangtang can structurally extract the features of various lesions from the massive image information generated by the digitization of CT, DR and other devices, so as to make auxiliary diagnosis for doctors. With the support of “AI big device”, this platform has rapidly expanded to cover more than ten key organs of the human body, covering the complete workflow of diagnosis, treatment and rehabilitation.
Our life is going through the process of “from virtual to real”, that is, after digitization, the content of virtual world is superimposed into real life through AI technology to create a new application model. For example, the AR navigation experience based on the Shangtang SenseMARS Mars hybrid reality platform has provided featured services in shopping centers, transportation hubs, scenic spots and other scenes such as Chengdu International Financial Center, Changsha Huanghua Airport, Hangzhou West Lake, etc.: users can open the app to scan the surrounding environment, and then get the ground arrows and virtual landmarks leading to the destination on the real-time screen.
Not only that, in the West Lake of Hangzhou, Shangtang also transforms the historical allusions of many scenic spots into virtual elements, bringing tourists an immersive experience through time and space. For example, in the Crane Pavilion, tourists can see AR plum blossoms and AR cranes. Even without a guide, they can feel the historical rhythm.
Smart cars are the epitome of “virtual and real integration”. With the support of SenseAuto in Shangtang, cars can turn reality into reality, structure road information, understand traffic conditions, and make correct driving decisions. They can also turn reality into reality, skillfully combine L4 automatic driving with AR technology, and create the first self-driving AR minibus, so that passengers can not only enjoy the intelligence and convenience brought by automatic driving technology, but also see the wonders of future cities through the AR window, It has created a new mode for the park and scenic spots.
The above scenarios are just the epitome of AI’s unprecedented breadth and depth in promoting social development and accelerating the upgrading of industrial structure. According to the White Paper on Core Technologies of Artificial Intelligence in China, AI has fully covered the basic elements of social operation and improved the overall operation efficiency. How to establish efficient AI infrastructure has become the key driving force for a new round of business innovation and competition.
In the past, machines played a more important role in recording digital information, which is convenient for us to query and make decisions. This “machine assisted” model defines our information age today. Now, AI technology has made machines more capable of insight analysis and decision-making judgment. It has defined a new era of “human-machine collaboration”. The real world will benefit from the linkage of virtual world to greatly improve efficiency, and the virtual world will enhance its innovative life experience through the enhancement of the real world. All industries and businesses will be reshaped.
“AI big device” is not only the barrier to Shangtang’s long-term competitiveness, but also the support of this long-term vision.