In February 2018, Wang Shuang, who had done data security research in the United States for more than 10 years, was stopped by an FBI official when he returned home.
The other party took out a new Chinese list of “overseas young high-level talents” just announced by China. The name of Wang Shuang happened to be in the list. He is the only expert in the field of medical privacy computing on the list.
Prior to this, he traveled back and forth between China and the United States several times without any hindrance. But this time, the FBI seemed to be particularly concerned about his identity as a “privacy computing expert.” After explaining to the other party for more than two hours, Wang Shuang was able to set off.
This seems to be a personal episode, but what is reflected behind it is more like a harbinger of industry development- in an era when data has become a factor of production, privacy computing is playing an increasingly important role and is being valued by major countries. This time Wang Shuang returned to China because he judged that privacy computing will usher in unprecedented development opportunities.
After that, it was exactly as he had predicted. First, on May 25, 2018, the European Union’s General Data Protection Regulation (GDPR) came into effect; one month later, California issued the most stringent privacy legislation in the United States, the California Consumer Privacy Act of 2018 (referred to as CCPA) to protect consumer privacy and data security. Since then, Facebook and Google have caused huge fines for forcing users to agree to share personal data.
After the law has imposed the strictest regulations on data security, privacy computing has become the best technical solution for the current data compliance flow. Since then, a number of companies engaged in privacy computing have emerged, including companies that have transformed from fields such as big data and blockchain, as well as subordinate departments of major manufacturers such as BAT, and specialized startup companies, including Wang Fengwei Technology founded by Shuang.
Especially this year, after a series of systems such as the “Data Security Law” and “Personal Information Protection Law” have been improved, domestic privacy computing has ushered in explosive development. According to the “2021 Privacy and Confidential Computing Blue Book” jointly compiled by China Mobile Communications Federation, Chinese Academy of Sciences, China Academy of Information and Communications Technology and other units, the data circulation market based on privacy computing can reach 100 billion yuan, of which medical, financial, and government affairs are Currently, there are three major application areas of privacy computing.
In August of this year, Fengwei Technology, which was founded by Wang Shuang and was established less than two years ago, successfully completed a billion-dollar B round of financing. “Two years ago, it would take a lot of time for investors to understand the concept of privacy computing, and this time the entire financing process only took two months.” Wang Shuang said. Behind the pursuit of investors is the acceleration of the entire privacy computing industry. This year, the revenue of Kaiwei Technology is expected to achieve a growth of more than 10 times. According to Wang Shuang’s optimistic forecast, the industry will have revenue of more than 10 times in 3-5 years. A billion-dollar private computing company.
This new computing paradigm, triggered by data security, is already accelerating on the front of science and technology.
Privacy computing, from concept to implementation
In 2016, Google AI first introduced the concept of “federated learning” in a technical blog post, and then this technology began to attract attention in the AI field.
The so-called “federal learning” is to allow multiple participants to build a universal and powerful machine learning model without sharing data.
To put it simply, the “data does not move and the model moves” method is used to solve problems such as data privacy, data security, data access permissions, and access to heterogeneous data. Since then, “federal learning” has gradually attracted attention, which has promoted the improvement of the entire privacy computing technology architecture.
But if it is investigated in detail, Google is not the originator of the core theory of “federal learning”, because in the previous four years, Chinese scholar Wang Shuang has discussed it many times.
At the end of 2011, Wang Shuang, who had just joined the University of California (UCSD) as a teaching assistant at San Diego, received a task from an academic perspective to build a privacy-preserving computing architecture serving the American biomedical computing network. This project is the core part of the National Biomedical Computing Center. It will use technology to open up the medical data of more than 300 hospitals in the United States and develop a set of standards.
At this time, there is no concept of “privacy computing” in the world, and all Wang Shuang’s work has to start from scratch. Fortunately, the doctoral research topic he just completed is distributed coding. This was originally part of cryptography. Wang Shuang tried to apply this technical concept to the new task of sharing medical data, and called it “secure federated learning.”
The main idea is that every time the data of each hospital is used, only the analysis model is put into the data pool for calculation, instead of extracting the data. This creative idea not only makes the data that the hospital has been sleeping for many years to play its value, but also solves the problem of data security.
In 2012, Wang Shuang published the world’s first online medical federation study paper in an SCI journal. This paper has also become the source of his academic research and work in the field of medical privacy computing in the future.
After completing the National Biomedical Computing Center project, Wang Shuang saw a huge demand for privacy and security in the medical field.
Those familiar with the medical industry may know that doctors, like university professors, have undertaken many scientific research projects in addition to their job of “rescuing the dying and healing the wounded.”
In theory, it is easier for doctors to abstract medical experience through their own clinical practice. However, the reality is that to abstract these experiences requires a lot of data verification, and the amount of data a doctor can get in a single hospital is far less than this requirement. For this reason, it is necessary to verify the experience data of doctors in multiple hospitals.
Medical informatization has been implemented for decades. Its original intention was to use technology to improve medical efficiency and break through the information barriers between hospitals. However, various privacy and security issues have always been between ideal and reality, making the medical informatization system established in the past practical. The above has only completed the task of improving efficiency internally, and the information exchange between the hospital and the hospital has not yet been realized.
There are difficult practical reasons for the incompatibility of data between hospitals. First, the patient’s data involves personal privacy and cannot be shared directly; second, the data exchange involves the data security of a hospital, so the hospital will not easily leak the data; third, the data is uncontrollable, and a hospital shares the data with cooperation Or, the other party may become a competitor after getting the data.
If the data security problem is not solved, the data sharing between hospitals cannot be fully realized under the current circumstances.
Wang Shuang found that using “federated learning” and other means to solve this problem from a technical level can achieve “data availability but not visible”, which greatly improves the usability of information technology in medical treatment. For example, in the past, hospitals were often limited by insufficient data in a single center when treating rare diseases. After solving the problem of data privacy, they could integrate industry-wide data to find the most effective treatment plan for patients. “It used to take several weeks to find a treatment plan, but now it may be shortened to 1 day or even shorter.” Wang Shuang said.
Caption: Wang Shuang, co-founder and chairman of Fengwei Technology, participates in the privacy computing discussion at the 2021 World Artificial Intelligence Conference
To truly implement this vision, technology and industry are required to run-in. The best way is to gather people from technology and industry through a type of activity to eliminate the “gap (gaps) between the cryptography group and the doctor group.”
In 2014, Wang Shuang initiated and organized the first iDASH secure computing competition with the support of the National Institutes of Health (NIH). Although only 10 teams came to the first competition, these teams are mostly college teams attracted by the sponsor’s “academic charm”, but after all, they have built a bridge of communication between the academic and medical industries of private computing.
Since then, with the continuous expansion of the scale and influence of the competition, the participating teams have expanded to hundreds of Internet companies and startups. Now the iDASH secure computing competition is one of the most influential events in the global privacy computing field, and it has become a competition field for major privacy computing vendors to show their strength.
What surprised Wang Shuang was that when geeks who mastered privacy computing technology communicated frequently with doctors in the hospital, the technology of privacy computing also had rapid iterations. “In terms of technical performance, there will be a 10 times improvement every year.” He has one. Obviously, it takes 1 hour to solve a multi-party joint modeling problem, but now it only takes 1 minute.
After breaking through the technological barrier, privacy computing is ushering in a storm of landing.
Old technology, new air outlet
In May 2018, the European Union officially implemented the GDPR, which is known as the most stringent data regulation in history. After that, it was rumored that companies such as Facebook and Google might face sky-high penalties. In the end, Google was fined 50 million euros by the French data protection regulator.
Taking this as a watershed, data security began to receive unprecedented attention from major technology companies.
At the same time, privacy computing has also become the focus of attention in the technology circle. Wang Shuang still remembers that in March 2018, he was invited by the Massachusetts Institute of Technology (MIT) to give a report on privacy computing. As a result, there were many Turing Award winners in the audience, as well as Chinese American scientists. , Stanford University professor Zhang Shousheng.
Immediately, a private computing entrepreneurial wave led by scientists and professors began to take off.
Beginning in 2018, Huakong Clearance, founded by Professor Xu Wei from the Institute of Interdisciplinary Information of Tsinghua University, Weiwei Technology, founded by Wang Shuang, a pioneer in the field of privacy computing and federal learning, and Light Tree and other entrepreneurs focusing on private computing services Companies have appeared one after another. At the same time, Internet giants such as BAT and previous companies in the fields of big data, blockchain, and AI have also stepped into or transformed into the field of privacy computing. After 2020, the development of the privacy computing industry will usher in the first round of climax.
“From the perspective of the degree of focus, the startup company will not appear to be disadvantaged in front of the big factory, but its neutrality is not possessed by the big factory.” Wang Shuang told “Jiazi Guangnian.”
Nevertheless, Wang Shuang emphasized that the technical threshold of privacy computing cannot be ignored. “After the first wave of private computing entrepreneurship, many companies began to package themselves as private computing companies based on the open source framework, claiming to the outside world that they had the ability to private computing, and then looking for investment institutions to get money.” He said.
On the one hand, the technology used in privacy computing is not a cutting-edge new technology. In addition to the federated learning mentioned above, there is also secure multi-party computing, which was proposed by Turing Award Chinese winner Yao Qizhi in the 1980s; in addition, there are trusted computing environments, homomorphic encryption, and differential privacy. Encryption technology proposed more than ten years ago. But when these technologies are applied, there will be performance tests. For example, whether the processing memory of encrypted data is too large, whether the running time of the system is 1 hour or 1 minute, and how accurate is the parameter construction of the model. “Without years of technology accumulation, this kind of optimization is impossible.” Wang Shuang said.
On the other hand, the application of private computing technology requires a deep understanding of the scene, which is the key to the commercialization of private computing. The reason is the same as the difficulty of AI landing in the past two years-it can only be achieved by cultivating the industry.
Based on these two advantages, Kaiwei Technology, established in October 2019, won the China Medical Information Big Data National Team’s project in two months and built a provincial-level medical cloud based on privacy computing with other participants. This privacy computing system can connect the data of hundreds of Grade-A hospitals, thousands of Grade-A hospitals, and tens of thousands of community hospitals under the jurisdiction of a province, and promote the value transformation of medical data.
Wang Shuang still remembers that when communicating with the relevant personnel of the Chinese Medical Information Big Data National Team, the other party paid particular attention to the “out of the box” of privacy computing. “They had contacted several startups in the industry before, but the final results were not ideal.” After communicating with Wang Shuang, the other party learned that he had done a project of the National Center for Biomedical Computing in the United States. With nearly 10 years of experience in the computing field, more importantly, the underlying technology platform of Fengwei Technology has been verified on tens of millions of data across the institute, which can be used “out of the box”.
Winning China’s medical information big data national team means that Kaiwei Technology first opened up the application of privacy computing in medical data networks in terms of breadth. But this layer touches more data similar to the “patient information home page”, lacking in depth.
For this reason, Fenwei Technology has found a project of the National Special Disease Network and reached a cooperation with it. The data covered by the specialized disease network is not as big as the provincial medical cloud, with only dozens of specialized disease hospitals at the head, but its advantage is that the data has sufficient depth. “Depth means that it has some fields formulated by experts. These fields are very helpful for the research of certain rare diseases and the research and development of new drugs by pharmaceutical companies.” Wang Shuang explained.
Based on the specialized disease network, Kaiwei Technology cooperated with the rheumatism immune network of a head hospital to deploy privacy computing technology to many top three hospitals under its specialized disease. Through the “data immobile model”, the original data “sufficiency” “Do not leave home” (do not leave the boundary of the data source), only transmit encrypted intermediate calculation results, and realize joint analysis across multiple hospitals. This achievement also won the first prize of Shanghai Municipal Science Progress Award.
At present, Fengwei Technology has completed more than 15 PoC (Verification Test) projects, and more than 50 potential customers.
Many industry insiders told “Jiazi Lightyear” that with the official implementation of the “Data Security Law” in China on September 1 this year, the implementation of privacy computing in the field of data security is also accelerating.
When will annual revenue be 1 billion?
Wang Shuang expects that this year, Fengwei Technology is expected to achieve 10 times revenue growth. He also has a bigger expectation that in the next 3 to 5 years, there will be companies with a revenue of 1 billion yuan in the privacy computing industry.
This is an exciting news for practitioners. But the real problem is how to commercialize private computing?
According to “Jiazi Lightyear”, in fact, most of the current business models of privacy computing companies still charge for projects and solutions. “Some seemingly high-revenue privacy computing companies are actually adding other business income, such as information systems, blockchain projects, etc.,” an industry insider said.
This has caused the industry to have doubts about the commercialization capabilities of private computing. Can private computing achieve large-scale revenue? How to make money?
According to Wang Shuang’s idea, in order to achieve the commercialization goal of 1 billion yuan, the profit model of private computing needs to be transformed into an underlying architecture based on private computing for development applications, and then revenue from related data can be made on this application. “It’s a bit like Didi and Meituan.” Wang Shuang said.
Based on past experience, Wang Shuang found that different industries and different scenarios have different requirements for data security, and their requirements for privacy calculations are different in efficiency and accuracy. Therefore, he and his team sorted out different industries and scenarios to abstract the core requirements; then added upper-level applications to the bottom platform, so that different applications on the platform can correspond to the needs of different industries and scenarios.
Legend: Fengweixin privacy computing platform and its modules
For example, by combining technologies such as federated learning, secure multi-party computing, and homomorphic encryption, a microservice can be formed. When a user puts forward a requirement, the corresponding service item can be found, and there is no need to re-customize the corresponding application based on each application scenario. .
This is equivalent to turning oneself into a data trading platform based on privacy calculations, realizing revenue through various trading applications on the platform. In this way, the commercial imagination of private computing will be infinitely magnified.
From the perspective of the industry structure, giant-based companies such as Ant Group and WeBank are all doing privacy calculations. However, “Ant itself is also a big party in data”. At this time, a neutral third party becomes important.
Gao Tianyao, a partner of Lenovo Star, said that the privacy computing field is still a two-party structure. “In the future, it will certainly gradually form a multi-party platform.”
To achieve this goal, the first step that needs to be solved is the problem of “data islands”. Privacy computing companies must first open up data within enterprise customers, and then achieve data interoperability within the industry.
In Wang Shuang’s view, this is the process of building private computing nodes. “In the past, many data sources did not have private computing nodes and could not provide external services. We deploy private computing client devices to the data source, which can realize the external services of the data source and derive more applications.” He said.
At present, this model has been applied in the domestic cancer special disease network. Kaiwei Technology deployed privacy computing to the special disease network, and built a cancer research data platform covering more than 60 hospitals in 24 provinces across the country. Then, pharmaceutical companies can do drug research and development analysis based on this platform, and insurance companies can also base on this platform. The platform is underwritten.
While building privacy computing nodes, Fenwei Technology is also promoting the establishment of industry standards, including working with 10 institutions such as the China Academy of Information and Communications Technology to promote the industry standard of “federal learning”, and participating in the development of national standards for privacy computing in the medical field of relevant ministries and commissions. Formulate.
With the gradual completion of privacy computing nodes and related industry standards, opening up cross-industry data has become a matter of course. By then, a private computing company with a revenue of 1 billion yuan will no longer be empty talk.
Posted by:CoinYuppie，Reprinted with attribution to:https://coinyuppie.com/how-can-a-privacy-computing-company-earn-one-billion-a-year/
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.