When I was young, I met a very funny and knowledgeable mathematics teacher, who told us many impressive stories about the history of mathematics, including the famous "Königsberg Seven Bridges Problem".
As an author who pays attention to the development of science and technology in China and the application of cutting-edge technologies, the most important domestic science and technology awards have always been the focus of my attention every year. The reason why I remembered this teacher and her story at this point in time came from a list of winners of the "2022 Chinese Institute of Electronics Science and Technology Award".
This award is approved by the National Science and Technology Awards Office, has been selected for 19 consecutive times, is the highest award in the field of electronic information in China, and has always been regarded as the vane of domestic electronic information technology development.
In the list of first prizes for scientific and technological progress announced this year, there is a project called "Intelligent Analysis Technology and Large-scale Application of Large-scale Complex Heterogeneous Map Data".
This project caught my attention for two main reasons. First, this project involves the hot "graph technology" in the scientific and technological industry, which is a cutting-edge technology that is very unfamiliar to the public, but has become a battleground for major giants in the scientific and technological industry, and its application has long penetrated into the daily life of ordinary people.
The second is this project, which was completed by teams such as Beijing University of Posts and Telecommunications and Ant Group. The scientific research cooperation model of universities + private enterprises represented by it has become more and more popular and valued in recent years, and it represents that Chinese private enterprises are growing and even growing into an extremely important force in the field of scientific and technological innovation in China.
Ant Group's research and development of graph technology is only a microcosm of this force, looking behind it, it may see the sea of stars for China's private scientific and technological innovation in the future.
Figure: A new way of understanding the world
In present-day Russia, there is an enclave called Kaliningrad, located between Poland and Lithuania.
Three hundred years ago, it was called Königsberg, and a river called Prigolia crossed the city and divided the city into two small islands in the center and the rest of the riverbank land, and the locals built seven bridges over the river to facilitate transportation.
Before the Soviet Union acquired it, it was one of the most famous cultural centers in eastern Germany, and it was also the "land of dragons" for the Teutonic Knights and Prussia. Countless tourists and pedestrians have walked through this busy and bustling city, and over time some people have asked a question when visiting this place:
How can a walker visit seven bridges at once without repeating or missing a visit, and finally return to the starting point?
The puzzle was never solved, so much so that it became a popular pastime for locals, with many claiming to have found a solution but failing when asked to follow the rules again. It wasn't until 1736 that the 29-year-old genius mathematician Euler wrote a paper called "The Problem of the Seven Bridges in Königsberg."
He regarded each piece of land as a "point" and the bridges connecting the land as "line segments", abstracted the German "land of dragons" into a simple and clear mathematical model, and distilled the problem of the seven bridges in Königsberg into a typical one-stroke drawing problem, and finally came to the accurate conclusion: it is a figure that cannot be completed in one stroke, and it is impossible for one to return to the starting point without repeating the road and visiting seven bridges at once.
Every time I think of this story, I feel the beauty and grandeur of time and the inheritance of human technology.
At that time, two hundred years before the birth of the world's first computer, no one would have dreamed that Euler's paper would open up the most important, interesting, and almost misunderstood basic theoretical system in the field of computer science in the future: graph theory.
No one will know that three hundred years later, graph technology based on graph theory, in today's era of unprecedented changes in a century, will become the most cutting-edge and most concerned outlet technology in the entire scientific and technological industry, and even become a place where soldiers must compete in the scientific and technological competition in the next hundred years.
According to Gartner's 2021 Big Data and Analytics Trends report, graph technology will be used for 80% of data and analytics innovation by 2025, between now and two years from now.
Today, everyone is familiar with the terms big data, AI, and algorithm recommendation, and almost all the applications you hear today that use big data analysis and AI algorithms, whether it is Alipay's mobile payment, or Douyin's algorithm recommendation, or accurate prediction of weather forecasts, complex fault analysis and guarantee of power grids, intelligent recommendations on major e-commerce platforms, and even you steal energy in ant forest...
By two years, at least 80% of it will be backed by graph technology.
Gartner, 2021 Big Data and Analytics Trends
The solution of the Königsberg Seven Bridges problem has become an important foreshadowing in the history of computer science development. In English, the picture is "image", the graph is called "graphics", and the "graph" of graph technology is in the same vein as the "graph" of graph theory, written as "Graph", which refers to a structural model, a system that deconstructs and looks at problems with "nodes" and "relationships".
In my opinion, this is a new perspective and a new system for understanding the world in the form of a "mind map".
The solution of the Königsberg Seven Bridges problem is the best interpretation of this new perspective. Euler treats each piece of land as a node and each bridge as an "edge" connecting these nodes, and this modeling method naturally focuses on "links" and "connections between things", and has better ability to describe the direction and properties of this connection.
Compared with the traditional way, understanding the world in the form of a "diagram" structure, especially the connection between things, is obviously more direct and advanced.
Take the relationship network of social platforms, I don't know if you have the experience of finding common friends in the watchlist of other friends, the traditional way is to list the friends of two people, and then look for the overlapping parts.
And the graph data itself is a data structure organized through the connection between things, so if you want to find "mutual friends", you don't even need to retrieve, because all relationships are completely clear on the graph, you only need to locate it.
After entering the era of big data with information explosion, artificial intelligence and machine learning began to emphasize correlation and causality more and more. The data construction method of "graph" is precisely concerned, and even based on correlation and causation. This data structure was almost naturally born for the AI era, and it makes many problems that seemed complex in the past easy to solve under this new perspective of understanding the world.
Figure the future of the industrial chain in technology
The most typical case of graph technology application is financial risk control.
When I did research on mergers and acquisitions in the primary and secondary markets in my early years, I had a job that I felt was very charming but also very bad, that is, to study equity relationships and capital chains, and a very important part of them was to study the guarantee chain.
For example, if Company A wants a loan, it says that I am looking for Company B for guarantee, Company B is looking for Company C, and Company C is looking for Company D. Company D said yes, but when you checked, you found that Company D may have asked B to make a guarantee. In the end, everyone formed a closed loop, and such a guarantee is simply invalid. Risk control is to find out such problems in the guarantee chain and avoid lending money out and not collecting it. If you understand it in the way of "diagram", it is actually to find and locate the "ring" part in the relationship network of guarantees.
The traditional approach is not impossible, but the time and resource requirements required to eliminate risks are high, and they are not as accurate and comprehensive, and the capacity is limited. For example, some banks use MapReduce big data processing methods, but in the enterprise guarantee scenario, it can probably only find a ring with a length of 6, and it cannot be found for a long time.
Compared with the guarantee scenario, the detection of cash transactions has higher requirements for timeliness, and not only has high timeliness requirements, but also needs to comprehensively consider a variety of information to improve the detection accuracy. Because although it is also necessary to identify the closed loop in the detection of cash transactions, the results obtained by simply detecting the closed loop of transactions are often not accurate enough.
For example, A transfers money to B, B transfers money to C, and C transfers money to A, which is a closed loop of transactions, meaning that there may be cash-out transactions.
But if A transfers money to B half a month, B transfers it to C; C will transfer money to A after another half a month. Although this also forms a closed loop, because black ash production crimes need to consider the time cost, this situation is often not cashing out in real life.
To more accurately determine the nature of such transactions, it is necessary to incorporate time information into the graph. This is called a "dynamic graph".
For example, in the same closed loop, if B is a large enterprise account, then the probability is that there is no problem with the transaction of A-> B. However, if A/B/C already know each other, and they are all transactions carried out through offline QR code payment, the possibility of cashing out is obviously high.
Incorporating information such as the type of node (whether it is a large business) and the type of edge (whether there is a social relationship) into the graph is the concept of the so-called "heterogeneous graph".
The traditional way of finding a closed loop itself is already difficult, not to mention that on the basis of finding a closed loop, other information needs to be considered comprehensively. So you have to do graph technology.
The graph data in real applications are complex, including both dynamic and heterogeneous, and these characteristics need to be modeled simultaneously to achieve higher recognition accuracy.
Ant began to study graph technology, initially driven by business needs. The highest number of transactions per second on Double 11 in 2020 is 580,000 transactions per second, and with such high concurrency, it is required to detect risks with great accuracy in a very short period of time, and to ensure that transactions can be completed on time, which is almost impossible in a traditional form. Therefore, some people have always said that China's mobile payment can be developed purely due to the high-frequency application scenarios brought by the demographic dividend, and the lack of technical content, I strongly disagree.
In the software world, the more easy-to-use, simple, yet fully functional the software, the more difficult it is to develop. The same is true in the field of payment, the more simple, fast, convenient and secure users can complete the payment, it means that the stronger the hard core technical strength behind it is needed.
Taking the award-winning project of "Intelligent Analysis Technology and Large-scale Application of Large-scale Complex Heterogeneous Graph Data", Beijing Post and Ant and other teams have worked hard for ten years, during which they have won 43 domestic and foreign invention patents, 51 CCF-A papers and other intellectual property achievements, and participated in the formulation of 2 industry standards.
The process is not without hardship, and of course the results are not fruitful.
According to official information, in the past two years, the technology project has saved more than a dozen enterprises in finance, Internet, communications, power and other industries about 700 million costs. As a practical example of the project, MYbank's "Dayan System" has increased the loan availability rate of small and micro merchants upstream and downstream of core enterprises from 30% to 80%. Alipay can now complete the risk determination of a transaction in 0.01 seconds, and the asset loss rate in 2021 is less than 0.98 per 100 million - that is, for every 100 million yuan of capital circulation on the platform, there is less than 1 yuan of loss risk.
The essence of the progress of human society is based on the improvement of productivity, and the essence of productivity improvement is the improvement of efficiency. The determination of a transaction twenty years ago may take 1 minute, twenty years later only 0.01 seconds, and the analysis and recognition of trillions of dynamic and heterogeneous data, which can support this efficiency and accuracy improvement, graph technology is undoubtedly an important basic technology to promote social progress and scientific and technological innovation in the future.
Its application scenarios are not only in the financial field, but also in the energy, transportation, and medical fields.
For example, weather forecasting, the principle is to divide the earth into more than 1 million small grids according to the "longitude-latitude-barosphere", and determine the weather status on the grid. Changes in the weather of a grid can be affected by distant meshes, and the difficulty in predicting weather lies in how to model the efficient and sparse interaction between such a large number of meshes.
Image source: Related papers
DeepMind, which has developed AlphaGo, and Google have newly developed a weather simulator based on machine learning, GraphCast, which models these small grids in the form of a multi-scale grid graph, using a 16-layer graph neural network to achieve local and global information propagation, which can capture longer spatial interactions than traditional methods, thereby greatly improving the accuracy of weather prediction.
Forecasts of weather trends over a period of 4 to 10 days, known as "medium-term weather forecasts", are accurate and critical to policymaking in sectors such as agriculture, construction, and tourism. Currently, GraphCast can predict the weather in the next 10 days in 60 seconds, and the accuracy is extremely high.
Large-scale manufacturing industries, such as the semiconductor industry and the new energy industry, are also important scenarios for the application of graph technology. These industries with long process routes, many design equipment, complex parameters, and a large amount of product data, often have a very complex correlation between each link and data, and graph technology is particularly suitable for dealing with complex relationships and discovering hidden special associations, which can not only analyze risks, but also predict potential equipment, processes, and supply chain risks. Wherever relevance is required, graph technology has the natural advantage of timeliness and accuracy.
Social media and e-commerce recommendation systems, which are extremely daily scenarios, have been deeply penetrated by graph technology, and graph technology has extremely broad application prospects at the level of new drug research and development, intelligent transportation, industrial Internet of Things, and digital upgrading of traditional industries.
The innovation vitality of private enterprises
The project "Intelligent Analysis Technology and Large-scale Application of Large-scale Complex Heterogeneous Map Data" is only a microcosm of the cooperation between Chinese private enterprises and universities to promote China's scientific and technological innovation.
According to 2020 data, the mainland's private economy contributes more than 50% of tax revenue, more than 60% of GDP, and more than 70% of technological innovation.
In 2021, the whole society in mainland China will invest 2.79 trillion yuan in R&D, of which 76% will come from enterprise investment; In 2021, 79% of the national key R&D plans were led or participated by enterprises, while the investment of private enterprises is still growing, and in 2021, the research and experimental development (R&D) expenditure of mainland enterprises was 2.15 trillion yuan, an increase of 15.2% over the previous year.
I deliberately pulled out the list of all the winners of the first prize of the Science and Technology Progress Award of the Chinese Institute of Electronics in the past ten years from 2012 to 2022. In the early years, most of the main completion units of the award-winning projects on this list were research institutes of major universities and related majors, and the private enterprises that appeared on the list were mainly some communication equipment hardware and service providers, such as ZTE, Huawei and so on.
And with the passage of time, more and more private enterprises from all walks of life began to appear on this list, Ali, Baidu, Tencent, Netease, Qianxin, OPPO, TCL, Hisense, BOE, Ant Group... These Internet giants and leading companies in their respective fields have begun to appear on the list.
According to the Report on the Number of Invention Patent Authorizations of Private Enterprises in China released by the Intellectual Property Development Research Center of the State Intellectual Property Office, by the end of 2021, the total R&D expenditure of the top 1,000 private enterprises in R&D investment had reached 1.08 trillion yuan, accounting for 38.58% of the national R&D investment and 50.16% of the national enterprise R&D expenditure; The year-on-year increase was 23.14%, 8.5 percentage points higher than the national growth rate and 7.9 percentage points higher than the national enterprises.
The total number of valid patents at home and abroad among China's top 500 private enterprises was 633922, an increase of 53.60% over the previous year. Among them, there were 134657 internationally valid patents, with an increase of 474.65%. In the past 20 years of development under the guidance of the Outline of the National Medium and Long-term Science and Technology Development Plan (2006-2020), more and more scientific and technological achievements from private enterprises are emerging.
Unknowingly, the global economic competition pattern has changed, and the comprehensive national strength competition between countries has evolved into development competition with scientific and technological innovation as the core.
Stuck in the neck is a very painful thing, and it has been buried in the heart of every Chinese all these years. But what makes people feel better is that such an international environment has further promoted the improvement of Chinese enterprises' awareness of independent innovation, forcing enterprises to embark on the stage of international scientific and technological competition. A large number of them are private enterprises.
Many previous technical gaps are being filled little by little, mobile phone operating system has Huawei's Hongmeng, artificial intelligence We also have Baidu's independent research and development of deep learning platform, cloud computing field Ali has deep technical accumulation, new energy vehicle field from BYD to Wei Xiaoli, have been at the forefront of the world. China's private enterprises are becoming the representatives of the mainland to participate in international scientific and technological competition and grasp the discourse power of international scientific and technological innovation, and become a force that cannot be ignored in the global market with unimaginable scientific and technological innovation vitality.
epilogue
In 2011, Professor Ishikawa of the Department of Computer Science of Beijing University of Posts and Telecommunications first came into contact with an important concept in graph intelligence technology, Heterogenous Graph.
This concept was first proposed by international data mining authorities Han Jiawei and Yu Shilun in 2009. At that time, the acceptance of the concept in the industry was still very low, and some of the foundational works in the field were rejected several times.
But good technology shines, and Ishikawa is not willing to be buried with such technology, and embarks on a preaching path that lasts for more than ten years. Later, Ishikawa met Ant Group, which has advanced mapping technology in the industry, and the two sides immediately hit it off to set up a scientific research group. After the test of large-scale industrial-grade scenarios, Ant has mastered graph technology with ease, and has combined with the academic sword, giving birth to the graph intelligence research results mentioned at the beginning, raising the level of graph intelligence to a new level.
Figure such technology, to a certain extent to develop a need for its own scale is large enough, application scenarios enough platform as a stage, in order to further evolution and iteration, this kind of thing small entrepreneurial teams are difficult to do, the past two decades of growth of China's technology industry giants, has become an important force for the large-scale promotion and application of cutting-edge technology.
And there are more enterprises, although not necessarily as powerful as large platforms, but also gradually grow into key technology enterprises in the industrial chain.
Since the Ministry of Industry and Information Technology issued the "Implementation Plan for the Special Action for the Cultivation and Improvement of Single Champion Enterprises in Manufacturing Industry" in 2016, a total of six batches of "single champions" have been announced, and these enterprises have achieved international leadership in production technology or process in the market segment, and more than 400 of the 848 "single champions" are private enterprises.
Last year, we counted China's aircraft carrier-related suppliers, and there were 532 supporting units of Shandong ships, of which 412 were non-military social supporting units, of which private enterprises accounted for half of the total.
Before I wrote this article, I just went to brush "The Wandering Earth 2", the director of the film, Guo Fan, is often ridiculed as the most "fate" director in China, when shooting the ball 1 with a grass team, the ball 2 has established a set of standardized management system that completely belongs to the Chinese film industry, realizing cloud 5G media asset management and live streaming of on-site shooting.
In about 5 years, the main creative team of "The Wandering Earth", which tried to make a way for Chinese science fiction films, changed from a grassroots team to a modern and industrialized film studio, and became a representative of China's advanced cultural productivity.
With the efforts of all these well-known and unknown companies, visible and invisible, China's private enterprises have become an important force in scientific research innovation and social progress in the mainland, and burst out beyond imagination of scientific and technological innovation.