There is no doubt that in the field of privacy computing, no enterprise can be so unanimously regarded as a “big brother” by peers as Huakong qingjiao, and Professor Xu Wei, the founder of Huakong qingjiao, is a benchmark in the industry.
Professor Xu Wei has always stressed that the leap of Huakong qingjiao from zero to one is inseparable from the efforts and efforts of the enterprise team. He said bluntly: “what a reliable team does is reliable, but unreliable people can do unreliable things even if they are unreliable.”
As an associate professor of the cross Information Research Institute of Tsinghua University and the founder of Huakong qingjiao, Professor Xu Wei is a practitioner of “industry university research”, which is deeply influenced by his mentor Professor David Patterson and academician Yao Qizhi of bole.
Professor Xu Wei recalled his experience of studying and returning home to start a business, which was almost emotional. Meeting academician Yao Qizhi also became a major driving force and career turning point for Professor Xu Wei to return to teach in 2013, just as Qianlima met bole. In 2018, Tsinghua University initiated the establishment of Huakong qingjiao by transforming the scientific research achievements of academician Yao Qizhi and Professor Xu Wei.
However, Professor Xu Wei also admitted that at the beginning of the establishment of CCCC, few people realized the importance of privacy computing. It is precisely in this way that his team has more time to precipitate the research and development of general-purpose privacy computing infrastructure, which has laid a solid foundation for Huakong clearing and payment, and is also ready for the popularization of privacy computing.
In 2021, CCCC participated in the establishment of Beijing international big data exchange and Hunan big data exchange, providing important technical support for China’s new data exchange.
According to Professor Xu Wei, big data exchange has experienced three generations in technology. The first generation is plaintext data exchange, and data trading is very limited; The second generation is API data transaction, which is only applicable to some specific statistical scenarios. The third generation of new data exchange is to introduce privacy computing technology and conduct ciphertext calculation based on multi-party data aggregation. It does not conduct direct transactions between data, but transactions of specific use right value of data.
“Privacy computing reduces the technical risk in data sharing, but data trading is a combination of technology and management, so data exchange is very necessary.” Professor Xu Wei warned.
With regard to the commercial landing of privacy computing technology, Professor Xu Wei mentioned that the commercial landing of privacy computing can achieve large-scale development in recent two years, but the real large-scale use still needs the real establishment of big data ecology, which needs more time to be completed, and Huakong qingjiao is participating in it.
Perhaps we can learn more about the future of Huakong clearing and privacy computing from Professor Xu Wei’s narration.
01 I like free exploration and innovation
Computational power think tank: from the outside world, your identity is diverse. Are you a professor, founder, scientist or winner of various awards? What advantages does your diverse identity bring to you?
Xu Wei: I think these identities complement each other, “industry, University and research” promote and help each other. I am a researcher of engineering discipline. The most important thing is not only to have theoretical research output, but also to be really applied in social production.
From the perspective of research, academic research can better explore what the definition is, what the core problem is and what methods can be used to solve it. For example, what problems does privacy computing want to solve, and what fields does it belong to. First understand the problem, and then come up with solutions from scientific research and cutting-edge disciplines, rather than using whatever methods I know.
From the perspective of education, the knowledge given to students is not only theoretical on paper, but also new discoveries in engineering practice and knowledge that can be truly produced in the industry.
From the perspective of enterprise production, team members perform their respective duties, including management, scientific research and business docking. Everyone does what they are best at, so that industry and scientific research can promote each other rather than dismantle each other.
Multiple identities enable me to think about privacy computing from different angles and integrate between “industry, University and research”.
Computational power think tank: learning from Professor David Patterson, the winner of the Turing prize, and working in the cross information research institute founded by Professor Yao Qizhi, the winner of the Turing prize, what have these two good teachers and friends brought to your research or entrepreneurship?
Xu Wei: the impact is very great. I am a typical person who has been changed by education. When I was an undergraduate, I was very introverted, but after the exercise during my PhD, I felt that I had gradually changed.
Professor Patterson has great influence in the combination of industry and industry, guiding and encouraging me to combine academia with industry. Mr. Yao also thinks so. He once said that “to be an interdisciplinary computer discipline is not only to be a tool discipline, but also to use computer methods and ideas to change the practice of all walks of life”.
“Wrap technology in production, life and real economy”. After the establishment of Huakong qingjiao, we also chose the R & D of general-purpose technical facilities, which can be used to greatly improve the data use efficiency and reduce the use cost of various industries, so as to prepare for the popularization of private computing in industrial applications.
When Huakong qingjiao was founded, I had graduated for many years. Later, I met my mentor, Professor Patterson, who also supported me. He said that there are both successes and failures in starting a company. Whether it is successful or not, it is a very valuable experience.
Computational power think tank: you worked at Google’s U.S. headquarters from 2010 to 2013. What is the opportunity and motivation for you to return home for development?
Xu Wei: I’ve always wanted to go home. At that time, there were two opportunities. The first was to do research on infrastructure reliability in Google. The work was very stable and free, but it felt a little similar to retirement, and I preferred to explore new things freely.
At that time, Mr. Yao Qizhi just went to the United States to recruit. After my mentor Patterson introduced us, Mr. Yao introduced to establish an information Cross College in Tsinghua. I think it is a good opportunity to join the exploration early. Mr. Yao told me that he was free to explore the direction of interest.
I always thought that I was “picked up” by Mr. Yao from the United States. After graduating from doctor’s degree and working at Google, people who walked out of the academic circle were like living on the streets. But Mr. Yao gave me the opportunity to return to the academic circle. I am very grateful.
After returning home at that time, I explored many directions. Around 2014, although big data was very popular, it could not bring business value, because plaintext data transactions would bring too many problems, including compliance, disclosure of corporate secrets and so on. After realizing the problem of data sharing, I found that there are opportunities in this field. Some large enterprises, such as Google, have a closed-loop data, which can be accessed to generate data, and it can be realized with these data in other (such as advertising) businesses to form a closed-loop. However, in most enterprises, there is no data closed loop. Either they do not generate data themselves, or the data cannot be realized. This is an important reason why the field of big data has not developed. I think private computing can make data flow controllable. Later, I focused on exploring the direction of private computing.
Computational power think tank: in recent years, what fields have you mainly studied in many articles published in top academic journals? How about the transformation of achievements?
Xu Wei: my thesis is diversified and has several research directions. For example, data center network (especially high-performance data center network), system reliability, medical AI, privacy computing (a comprehensive field with multiple technologies), blockchain and other major fields. The diversity of research directions is the characteristic of professors in many systems fields.
In these directions, I chose to set up an enterprise to implement the research results of privacy computing. An important reason is that it is never just a cryptography problem. The improvement of privacy computing efficiency is also combined with a large number of AI model training methods, such as federated learning. These optimizations will also reduce security, so trade-offs are needed. Privacy computing also involves network transmission, LAN and WAN, which affect the speed of data access, hardware acceleration and simple and clear programming language. It needs to be considered from a more comprehensive and complete system perspective. That’s what I like.
Other areas of my research have also been transformed, but I have not personally participated in the industrial landing.
Computational power think tank: Bastille, an American company specializing in the transformation of technological achievements, once conducted a survey: the failure rate of entrepreneurship by professors in American universities has reached 96% ~ 97%. However, at a time when science and technology is the primary productive force, scientists’ entrepreneurship has also brought the most advanced achievements and ideas to the business field, Cambricon Technologies Corporation Limited(688256) , Shangtang technology, horizon and other enterprises are very famous examples. Huakong qingjiao has also become a leading enterprise in the industry. What qualities do “scientist entrepreneurship” need? What suggestions do you have for scientists, professors and entrepreneurs?
Xu Wei: I hope not to label professors and scientists. In fact, anyone who starts a business has success and failure.
It is very important for entrepreneurs to recognize what they are good at. Many scientists and professors think he can do anything. There is a reason for this perception. In the academic field, professors need to operate their own laboratories, from personnel to finance to talking about projects, and even painting walls and connecting network cables. In an enterprise, the efficiency and effect will be better if it is handed over to professionals. Professor entrepreneurship, I always remind myself that the most important point is open mind. Embrace change and different perspectives. If you are not open-minded enough and stick to your own cognition, it is difficult to find a suitable team to do things well.
I have known CEO Zhang Xudong for a long time in China. We trust each other. I am responsible for technology research and transformation. He is responsible for enterprise management. He cooperates very tacitly and complements each other. In addition, he will find some capable people to work harmoniously. Therefore, it is also crucial to establish business partners and teams.
If the team, business and other things have not been figured out clearly and are not mature enough, the professor can also turn over the results to others to do. Many scientists around me have adopted this method, and the transformation of achievements is also very successful.
02 construction of “new” data exchange
Computational power think tank: looking back, in 2021, in which areas did CCCC make more outstanding achievements in privacy computing? What are the plans to make breakthroughs in 2022?
Xu Wei: in 2021, outstanding achievements were made in the financial field and data exchange. Our technical feature is “universal technology”, and we are trying in various fields.
General technology can quickly form different algorithms and customized applications suitable for different fields, and we don’t focus on which field to develop. Universal privacy computing is the idea of a single system and distributed system, which supports different privacy computing protocols. As a tool, privacy computing technology is becoming more and more popular. Reducing the practical threshold is to create the technology of “rotten Street” and reduce the development cost. Turning a technology into a “rotten Street” is not only the ultimate pursuit of software system, but also the secret of the rapid development of computer industry, and privacy computing is no exception.
In 2021, in the technical field, we continued the overall transformation and R & D of the general-purpose system, so as to greatly reduce the use threshold and reduce the development and use cost of industrial applications.
In 2022, the plan in the technical field is to further promote the application of the system, so that manufacturers who want to do private computing can develop their own private computing products, and the integrator’s own technicians can complete the construction, so we don’t need to participate in it; Then, from the application of large-scale systems to the promotion of large and small-scale systems, before we deployed large enterprises, we carried out hardware acceleration optimization scheme to make it miniaturized and further popularize the use of private computing; The last is to provide more diversified security and performance options, including reducing costs through federated learning and other technologies in scenes that do not emphasize security, and improving the performance of ciphertext algorithm in scenes that emphasize security.
Overall, our goal is to enable more enterprises to use private computing, build infrastructure on a large scale and serve more enterprises. Even enterprises that do not want to build infrastructure can use cheaper and more efficient private computing services.
Computational power think tank: in 2021, Beijing international big data exchange and Shanghai data exchange have been established successively, Shenzhen data exchange is also preparing, and Huakong clearing house has also participated in the establishment of Beijing stock exchange and Hunan big data exchange. What kind of “new” data exchange is the goal of Huakong clearing house?
Xu Wei: making the data available invisible, controllable and measurable can really trade the specific value of the data. This is the new data exchange.
We have participated in the design of technical facilities of Beijing international big data exchange and Hunan big data exchange, and are technology providers.
The big data exchange has experienced three generations in technology. The first generation is the plaintext data exchange, and the data trading is very limited, and almost no enterprises trade in it; The second generation is API data transaction, which is only applicable to some specific statistical scenario analysis. After the data transaction, the buyer can calculate the data. The third generation of new data exchange is to introduce privacy computing technology and conduct ciphertext calculation based on multi-party data aggregation. It does not conduct direct transactions between data, but transactions with specific value of data. It is a new type of contract. After providing data, what algorithm is used to calculate and who owns the result, which is the content of the transaction.
The data trading platform established by privacy computing technology is a complex of technology and management. It is not the kind of public chain that only trusts technology and does not need management at all. In the operation of data exchange, technology solves technical risks. Management risks, authenticity of data sources and installation operation level still need to be guaranteed through management, supervision and audit provided by all parties. Therefore, the form of data exchange is necessary, and technology companies will not replace data exchange.
Computational power think tank: is the launch of China Everbright Bank Company Limited Co.Ltd(601818) enterprise level multi-party secure computing platform a milestone for the settlement of China control? Why? What is the specific operating principle?
Xu Wei: from the perspective of landing, this is the first complete enterprise level open source framework handed over by CCCC. It is also the first enterprise level open source framework used in production in the financial industry. It is a milestone.
As an enterprise level data circulation infrastructure, China Everbright Bank Company Limited Co.Ltd(601818) multi-party secure computing platform has the following characteristics: universality. The platform integrates a variety of privacy computing technologies such as secret sharing, homomorphic encryption, inadvertent transmission and federated learning, which can meet any algorithm requirements; Scalability: distributed technology architecture is adopted, data, algorithm, computing power and control surface are decoupled layer by layer, and scheduling system, computing engine and data service can be flexibly expanded.
High performance, ten million level data, minute level joint modeling, second level joint statistics and stealth query, which can be smoothly extended to multi-party secure computing of 100 million level data; High availability, local cross machine room load balancing, dual active deployment, machine room and server failure, automatic and seamless business switching.
Of course, from a technical point of view, this system is not the most complex for us. It is the same as the technical scheme we have always deployed.
Computational power think tank: in terms of technology, financing and implementation results, CCCC has become a leading enterprise in the field of private computing outside China. How does CCCC become the top in the industry?
Xu Wei: I think team reliability is the most important. What a reliable team does is reliable, but unreliable people can’t do anything reliable.
In addition, the ability to complete the construction of general-purpose technology can also be traced back to the beginning of entrepreneurship. At that time, users, perhaps including ourselves, could not see the needs of privacy computing. At that time, we had to build general-purpose technical facilities to meet the needs of the “future”. This later became a first mover advantage for us.
Now the technology of private computing is very popular. If it takes a few years to build a general-purpose technical facility, the enterprise may spend a lot of time and cost. New enterprises joining this industry, starting from a specific application, are actually suitable for the current stage of development.
Computational power think tank: how much is the latest valuation now? In the medium and long term, what is the development goal of Huakong qingjiao?
Xu Wei: on October 13, 2021, Huakong clearing and delivery completed the round B financing of RMB 500 million. The old shareholder Lenovo venture capital continued to invest and increase its holdings, Beijing Centergate Technologies (Holding) Co.Ltd(000931) Science City, oppo group, xunze technology, China International Capital Corporation Limited(601995) , Shanghai Pudong Development Bank Co.Ltd(600000) Puxin capital, Huaxing capital, longmafeng capital and Tongchuang Weiye jointly invested. The post investment valuation of round B financing of China Holdings clearing and payment exceeds RMB 4 billion; In 2022, it is expected to meet the valuation standard of Unicorn.
The medium and long-term development goal of Huakong qingjiao is to improve the data ecology, become the world’s top enterprise, “revive” the big data industry, and realize the closed-loop data sharing among more different enterprises.
Suanli think tank: in terms of technical concept, what safety assumptions does Huakong follow in clearing and delivery? Can we briefly introduce the new data concept and new data security concept we advocate?
Xu Wei: the security assumption is the choice made by the customer according to his data and application scenarios, not the choice of our technology provider. Technology can provide security and reduce data use risks, but there are other non-technical risks, among which the correctness of security assumptions is very important.
What are the safety assumptions? Different scenarios require different security assumptions. You can completely believe in human nature or pure code. The security assumption I recommend is “secret sharing”. I believe there is no collusion among multiple participants. Similar to “let the power run in the sun” and “operator system”, it is combined with a perfect supervision mechanism under the technical guarantee.
In short, security assumption is a trade-off between cost and risk.
When it comes to the concept of data security, the first point is that data cannot be circulated directly. Data is high-dimensional. Not only personal identity information needs to be protected, but also statistical data related to national security and enterprise secrets are sensitive. However, the algorithm is constantly developing. Once there is a more advanced algorithm, it is unknown what the data can do and what problems can be caused. Therefore, no one dares to directly circulate data.
Second, data should not discuss attribution. Data is more about custody responsibility, so it is difficult to discuss ownership.
Third, the “controllable and measurable” use of data is the core issue of data security and the purpose of adopting all these technologies. “Available invisible” is only the basis and means to ensure the controllable use of data.
03 the improvement of big data ecology is the premise of large-scale application of privacy computing
Computational power think tank: after data is defined as a factor of production, on the one hand, it brings new opportunities to enterprises with a large amount of data, on the other hand, it also brings more stringent supervision. What role does privacy computing play?
Xu Wei: privacy computing is expected to become the infrastructure for the circulation of production factors, break the data monopoly and change the inefficient production mode of “selecting data for production”, so it can improve the production efficiency and benefits of the whole society.
Computational power think tank: the business model of privacy computing is relatively single. Which model currently accounts for the main revenue structure of privacy computing enterprises? What is the current business revenue capacity of privacy computing?
Xu Wei: privacy computing business is a business model that provides software and hardware technology and core technology providers. However, after the ecological development of big data becomes active, socialized big data can circulate safely, and more business models will emerge.
At present, we focus on how to be a good provider of core technology.
At present, the business revenue capacity of the privacy computing industry is far from reaching the level we want, but this situation needs to be changed. What we need to do is a technical solution that can really bring value to customers and focus on tob business. The industry will develop. From the perspective of clearing and payment of China Holdings, we have taken a big step in terms of revenue in 2021, and will continue to maintain high-speed growth in 2022.
I think there will be an explosion of private computing in the next few years. At that time, people will not regard private computing as a mysterious black technology, but as a part of the big data ecology. At that time, the industry will really take shape. At that time, the enterprises in the industry will form a new business model..
Computational power think tank: in addition to privacy computing, which industries will usher in the “golden decade” in the next decade? What do you think of the development of the meta universe?
Xu Wei: I am optimistic about AI technologies for specific fields. These technologies can solve the problem that AI engineering has not kept pace with the current landing. There will be great prospects in the future in combination with different application scenarios; In addition, programmable accelerators for specific scenarios have great development prospects, and the system design combining software and hardware will become an effective idea to solve many computing problems.
Yuancosmos has directly brought UGC (user generated content) from microblog, short video and live broadcast to a new stage, allowing low-cost implementation of UGC with stronger interactivity and stickiness VR is only the external manifestation of the meta universe. In fact, the meta universe is also an interdisciplinary field. Yuancosmos has potential in breaking the monopoly of Internet platforms. However, the yuan universe is burning too fast, lacks the time for technical precipitation, and has an unstable foundation. It is prone to the industry risk that bad money expels good money, which will hit the whole industry. This point needs attention.