Not long ago, we took the 2021 East Lake International Artificial Intelligence Summit Forum as an opportunity to report that the "Chinese system" of artificial intelligence has begun to take shape in Wuhan, Hubei Province. I believe that many people have felt the exciting changes in the artificial intelligence industry from computing infrastructure, strategic planning, and scientific research promotion.
However, the end point of all changes must be enabled to the national economy and people's livelihood. Simply put, how does tall technology and infrastructure change the work and life we take for granted? Landing applications and supporting the digital and intelligent transformation and upgrading of thousands of industries is the only way for industrial AI and AI industrialization, and it is also a barrier that CANnot be bypassed in the planning, construction and operation of AI infrastructure.
In the past year, we have witnessed the Wuhan model of "one center and four platforms", marveled at the speed of Wuhan with a core construction cycle of only 5 months, so in terms of landing applications, the height of Wuhan that promotes the high-quality development of Hubei's digital economy with artificial intelligence is obviously worth seriously exploring.
At the end of 2021, let us see what kind of beautiful new world Wuhan has opened up a beautiful new world through the real changes in the AI industry that have really occurred in Jingchu Land.
Unlocking the shackles of computing power: pushing open the door to resources
Why the digital transformation and intelligent upgrading of thousands of industries should be premised on the construction of large-scale computing power infrastructure. This is to mention a trend, perhaps many people think that today's Chinese intelligent industry is very leading in the world, which is a fact, but there are still many shackles in the development of industrial intelligence.
Academician Zheng Weimin of Tsinghua University mentioned in his speech at the forum that computing power has seriously restricted the development of China's AI field. Especially in the integration of AI and science applications, artificial intelligence is rapidly moving towards the development of larger models, the more problem parameters processed, the better the processing effect, every 3-4 months the required computing power will double.
Therefore, while accelerating the integration of artificial intelligence with the real economy and driving the digital transformation and intelligent upgrading of the industry, constantly unlocking the shackles of computing power has become a problem that must be preemptively considered and laid out.
At present, more than 20 cities in China are planning or building artificial intelligence computing centers. Among them, the Wuhan Artificial Intelligence Computing Center has become the first artificial intelligence computing infrastructure with the nature of public service in China, with a typical benchmarking demonstration effect.
We note that after the Wuhan Artificial Intelligence Computing Center was put into operation, relying on abundant computing power, the Institute of Automation of the Chinese Academy of Sciences trained the world's first three-mode large model "Zidong. Wuhan University has trained the world's first deep learning framework for intelligent interpretation of remote sensing images "Wuhan. LuojiaNet”。 At present, it has achieved an average daily utilization rate of 92%, and more than 100 enterprises and scientific research institutes have settled in and used it, entering a period of steady rise.
There is no doubt that ai, big data, computing and other technology communities are becoming a strong driving force for the rapid economic take-off of Hubei and Wuhan, and a key engine for accelerating digital transformation. On the one hand, all industries and enterprises in Hubei need to introduce artificial intelligence to complete digital and intelligent transformation to achieve productivity changes to improve quality and efficiency; at the same time, using artificial intelligence as the base of innovation and supporting business innovation through new ICT infrastructure will help enterprises seize the technological dividend and shape the competitiveness of long-term development.
With the hashrate infrastructure of Wuhan, Shenzhen, Xi'an, Nanjing, Zhongyuan, Chengdu and other places successively online, the computing power infrastructure of more than 20 cities across the country will be connected to the network in the future, forming a national artificial intelligence computing power network. The hub position occupied by the Wuhan Artificial Intelligence Computing Center will also become the "nine provinces thoroughfare" of the digital world, as one of the core nodes of the efficient, advanced and green computing power network, helping Hubei and even the entire Chinese intelligent industry to push open the door of abundant computing resources.
Bearings in large models: pushing the door to efficiency
If you want to promote the application of AI, computing power is power energy, and the big model is the bearing that connects technology and industry.
Ai to transform into a productivity base for all walks of life, enterprises and developers need to complete the iteration of algorithm, software and hardware development in the subdivision field. If you want to customize the development for each vertical scene, not only is the cycle time long, but also limited by the real-world conditions such as data set, data scale, and development capabilities, and the model performance is likely to be less than ideal.
Therefore, the exploration of large models has begun all over the world, which is equivalent to the pre-positioning of a large amount of data clarity, data enhancement, model adaptation, etc., to technology giants or scientific research institutes to complete. By cooking a good "basic base material", developers in various industries only need to throw industry data "ingredients", throw them into the big model for training according to the standardized process, and do a small amount of scene adaptation, so that they can quickly develop high-precision industry AI models and create an "AI taste" of thousands of industries.
In May this year, Pengcheng Laboratory released the world's first 200 billion dense parameters of the Chinese NLP large model Pengcheng. Subsequently, the Institute of Automation of the Chinese Academy of Sciences incubated the world's first three-mode large model Zidong based on the Wuhan Artificial Intelligence Computing Center. Too early.
If it cannot be applied by scenes and industries, then the big model is just a display in the laboratory. Therefore, through the coupling of computing power and scientific research, Wuhan, Hubei Province, promotes the transformation and upgrading of local industrial clusters through the large model, which has played a certain reference significance.
First, release academic energy and incubate key models.
Hubei has 81 academicians, 130 colleges and universities, 1.6 million college students, more than 10 academicians leading innovation, strong scientific research capabilities, combining AI computing power with scientific tracks, and incubating a large model that can be used and applied in a targeted manner.
For example, the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing of Wuhan University requires a lot of manual work when extracting various remote sensing image information. After undertaking the project of "Large-scale Remote Sensing Image Sample Library Construction and Open Source Remote Sensing Deep Network Framework Model Research", Wuhan University cooperated with Huawei to build a sample bank and framework in the field of remote sensing imaging, relying on the surging computing power of Wuhan Artificial Intelligence Computing Center, and trained the world's first special framework for intelligent interpretation of remote sensing images for deep learning (Wuhan. LuojiaNet）。
Second, give full play to industrial advantages and promote industry innovation.
Hubei has more than 300 artificial intelligence enterprises such as Huawei and CITIC, and the international leaders in the fields of robotics, unmanned driving, and intelligent security have laid the foundation for the industry innovation of the big model.
With the world's first large model of graph, text and sound, Zidong. For example, Taichu can achieve efficient collaboration between the three modes of vision, text, and speech, and support the efficient completion of downstream tasks such as cross-modal detection, visual question and answer, semantic description, and video understanding and generation. At present, it has cooperated with iQIYI, Weiqiao Pioneering Group, Xinhua News Agency Technology Bureau and other enterprises to carry out multi-modal project cooperation, and try to apply it in industrial quality inspection, film and television creation, smart medical care, Internet recommendation, intelligent driving and other scenarios of national economy and people's livelihood.
In addition, we will establish an alliance ecology and create a bridge for industrial convergence.
After the leading large model tools are created by scientific research institutes, it is necessary to further converge the developer ecosystem in order to make the AI industry enter a virtuous circle. With the Wuhan Artificial Intelligence Computing Center as the base, after the birth of the big model, it further relied on the center to complete the transformation of production, education and research, empowered industrial applications with advanced scientific research results, and gathered two industries of intelligent remote sensing and multi-mode for Hubei.
At the 2021 East Lake International Artificial Intelligence Summit Forum, organizations such as the Intelligent Remote Sensing Open Source Ecological Alliance, the Multimodal Artificial Intelligence Industry Alliance, and the Wuhan Artificial Intelligence Research Institute were established, which means that the integration of industry, education, research and application with large computing power incubation of large models and "scientific research and innovation to accelerate the incubation of industry application innovation" is showing real and feasible persuasion on the jingchu land.
Two large model-related industrial alliances were established to attract professionals from other relevant domestic scientific research institutes, colleges and universities and enterprises, jointly build a scientific research ecosystem of intelligent remote sensing and multi-modal innovation in China, and promote the optimization of large models and the landing of industry applications.
Wang Jinqiao, a researcher at the Institute of Automation of the Chinese Academy of Sciences, revealed that members of the multi-modal alliance, such as art creation, news organizations or autonomous driving enterprises, have high requirements for computing power and models, and are more suitable for using the country's artificial intelligence computing centers and multi-modal large models and fine-tuned small models to do scenario-based applications through the alliance.
The big model of computing power enabling is promoting AI innovation in thousands of industries, reducing the cost and threshold of AI development, and making the artificial intelligence Chinese system run violently on top of the bearing.
Innovative application incubation: push open the door to the landing
With large computing power and large models, can we usher in the large-scale landing of artificial intelligence? From technology to industry, there is also a need for an ecological bridge.
At the 2021 East Lake International Artificial Intelligence Summit Forum, two major industrial alliances were established, and how to accelerate the incubation of innovative applications after the industry convergence has three solution references:
Solution 1: The big model is open source and open, creating a vertical model suitable for the industry.
Ascend Ecological Open Source and Open Concept has been unanimously supported by Pengcheng Laboratory, Wuhan University and the Institute of Automation of the Chinese Academy of Sciences, including Pangu, Zidong Taichu and Wuhan. Luojia et al. hold five large models, all for the global open source, interested enterprises and developers can download the corresponding models and datasets in the Ascend community.
Through the open source and opening of the big model, the influence of the industry alliance can be expanded, more innovative wisdom can be encouraged to join in, and the intelligent upgrading of various industries can be accelerated.
Solution 2: Exchange and share within the alliance to accelerate the incubation of industrial solutions.
As a bridge, the alliance has built a communication platform between the government, users, enterprises, scientific research institutes, and colleges and universities, and the various entities of industry, education, research and use can further connect supply and demand, share knowledge, and complement each other's advantages, creating a benign and efficient industrial development environment. Taking the Intelligent Remote Sensing Industry Alliance as an example, the world's leading remote sensing universities and remote sensing leading enterprises have joined forces to form a high-quality industrial chain convergence.
Among them, the Intelligent Remote Sensing Open Source Ecological Alliance, led by Academician Gong Jianya of the Chinese Academy of Sciences, has attracted leading enterprises, colleges and universities, and scientific research institutes from the field of intelligent remote sensing, such as Aerospace Hongtu, Space Technology Application in the 21st Century, Beijing Feidu Network Technology, and Beijing Supermap Software. Led by the Institute of Automation of the Chinese Academy of Sciences, the Multimodal Artificial Intelligence Industry Alliance has attracted many enterprises, institutions and research institutes such as China Mobile, Xinhua News Agency Technology Bureau, Huawei, and iQiyi to join.
Solution 3: Accelerate the landing of AI applications with the help of Ascend "Smart" series of solutions.
For some enterprises that want to directly apply AI and have limited development capabilities, Huawei has successively released four major solutions for Ascend Smart Manufacturing, Ascend Smart City, Ascend Smart City, and Ascend Smart Tour, covering manufacturing, smart city, transportation, energy, and other fields, so that end customers can directly copy and use mature AI scenario-based solutions to promote intelligent industry upgrades.
Wisdom will become a city, and AI will flourish. It is not difficult to see from the above scheme that the Wuhan Artificial Intelligence Computing Center and the Ecological Innovation Center truly and effectively support the landing of industrial AI and AI industry, and as the doors of computing power, large models and applications are pushed open one by one, the computing power network is rapidly realizing value release.
The moment of AI awakening: blowing the future rallying call from Jingchu
Hubei sounded the industrial rallying call, which is the starting point for the acceleration and upgrading of AI industrialization and industrial AI. What is more worth looking forward to is that more and more AI infrastructure is growing on the land of China.
On December 29, Nanjing Artificial Intelligence Computing Center was officially put into operation, combining Nanjing's scientific research advantages and industrial positioning to explore the "Nanjing model" of artificial intelligence. On the same day, at the Jiangsu Ascend Artificial Intelligence Ecology Conference, Nanjing Artificial Intelligence Computing Center signed a contract with Southeast University, Nanjing University of Science and Technology, Jiangsu Institute of Schistosomiasis Prevention and Control, etc., and jointly released the "Jinling. Electromagnetic Brain """ Jinling. Nezha ""Jinling. Naturalist" big model.
In addition, a total of 10 enterprises, including Jiangsu Hongcheng Big Data Technology and Application Research Institute, Nanjing Yunchuang Big Data Technology Co., Ltd., and Nanjing Anyuan Technology Co., Ltd., signed on-site contracts with Nanjing Artificial Intelligence Computing Center for computing power cooperation.
As Xu Yingtong, president of Huawei's Ascend Computing Business, said, as more and more artificial intelligence computing centers are launched, there will be more original technological achievements in different fields in the future.
From the "Wuhan model" to the "China model", the grand story of the artificial intelligence computing center has just begun. However, from a series of construction, research and development, and landing processes in 2021, it is not difficult to get some important enlightenments about the development of the AI industry:
First of all, the combination of industrial application and scientific research and innovation. Both are CV-oriented Pengcheng. Dasheng, Wuhan for intelligent remote sensing. Luojia, Jinling for electromagnetism. Electromagnetic Brain, as well as Pengcheng. Shen Nong, Zi Dong. Too primitive, the incubation of these large models is not a closed door in the laboratory, but the result of industrial application-oriented, industry-university-research-use active participation, and it is precisely because of the application of learning that it can truly bring together the industry and benefit the industry.
Second, the synchronization of computing power construction and ecological construction. From the perspective of the Wuhan model, the smooth progress of AI industrialization and industrial AI requires the joint efforts of multiple forces, and infrastructure construction should be promoted simultaneously with industrial ecological incubation. Wang Tao, managing director of Huawei and director of the ICT Infrastructure Business Management Committee, also pointed out that the successful practice in Wuhan shows that the development of the artificial intelligence industry is inseparable from the cooperation of all parties in the "government, industry, academia, research and application".
On December 19, 2021, at the main forum of the "2021 New Generation Artificial Intelligence Academician Summit Forum and Qizhi Developer Conference", Mei Jianping, deputy director of the High-tech Department of the Ministry of Science and Technology, Gao Wen, director of Pengcheng Laboratory, Academician Wang Huaimin of the Chinese Academy of Sciences, Professor Huang Tiejun, president of Beijing Zhiyuan Artificial Intelligence Research Institute, and Zhu Zhaosheng, general manager of Huawei Cluster Computing, jointly launched the establishment ceremony of the "Artificial Intelligence Computing Network Promotion Alliance".
It is reasonable to speculate that this "one center and four platforms" artificial intelligence computing center construction model will become the strategic focus of the development of the AI industry in the next period of time, accompanied by the accelerated formation of the computing power network.
The Wuhan model can be generally recognized by people in the fields of government, industry, academia, research and application, precisely because it has pushed open the doors of AI landing applications, allowing the public to see the magnificence of the new world of intelligence, and feeling the logical self-consistency, achievement incubation and value improvement brought by the artificial intelligence computing center.
This is why AI can awaken in the here, now, and here.