Author: Qiu Xiangyu
On December 5th, 2020 World Blockchain Conference Wuhan officially opened at Wuhan International Convention and Exhibition Center. The conference was hosted by Babbitt and received strong support from Wuhan Municipal Government, Jianghan District Government, Wuhan Municipal Bureau of Economics and Information Technology, and China Academy of Information and Communications Technology.
At the “Federal Learning and Private Computing Blockchain Business Innovation Development Application” roundtable forum held in the afternoon, Blue Elephant CEO Xu Min, Insight Technology founder and chairman Yao Ming, WeBank system architect Zeng Jice, and Guangzhishu Vice President Wu Shanshan and Fushu Technology Product Director Lin Lin jointly discussed the commercial innovation of federal learning and privacy computing blockchain.
How to view the game and balance between data commercialization and privacy?
Xu Min:
Data privacy and the commercial value of data are a pair of contradictions. Just like two waves photographed from opposite sides, they are rushing towards each other today. No matter how big the waves are, they will merge and move forward in the next second. I think Data privacy and data business value synergy are the same. Looking at it today, if we want to achieve data privacy protection and data business value synergy, I think there are three aspects, including humanities, law, and technology. Humanities, we will better achieve a balance between personal enjoyment of better social welfare, social services, and various privacy protections. The legal aspect is to determine the bottom line of data. Humanity and law are together, one determines the online, and the other determines the offline, and technology allows us to better achieve this balance.
Today we sacrifice 40% of privacy and achieve 60% of personal convenience. With privacy computing protection, we may only need to sacrifice 20% of privacy to achieve 80% of personal convenience. Privacy protects it It is to be able to improve social welfare.
How to grasp the integration of smart data and blockchain under the premise of data privacy protection?
Yao Ming: Let’s start with credit itself. The credit industry is a relatively special industry. The nature of the industry is actually using data and human historical experience, that is, models and wisdom to judge the risks of transaction entities. In such a field, there is a third-party credit evaluation scenario. In this scenario, there is a very important principle, that is, the principle of least available . If everyone is concerned about the credit industry, this term is frequently mentioned in policies and regulations. The least-available principle is actually to protect the subject’s privacy from being leaked and not infringed. This is also reflected in this year’s Civil Code. The other part is also to ensure fairness. For the least available, in fact, it realizes the visible part of the information as far as possible to prevent it from leaking out or being resold. For this minimum usable principle, if we define it as a protection for the visible part of information, technology now provides an opportunity to update. For those invisible calculated value parts, if we can extract it, yes Isn’t it possible to break this least usable principle, which is of great significance to credit evaluation, especially risk control.
Because from the perspective of risk control, it actually does not want to use the principle of minimum availability. He hopes to use more data to characterize the risks of the transaction objects, so as to quantitatively price credit and control risks. Both of these are actually The above is conflicting. Then this conflict provides opportunities for federated learning and multi-party secure computing. For data, the value of computing is extracted from its visible information, and this invisible computing value is removed as much as possible. Participate in sharing, participation in circulation, and participation in credit evaluation, so that you can break through the principle of least available from the technical level, but fully protect the visible part of personal privacy information and not be infringed or leaked. This is actually a technology for the credit industry Innovation.
How to verify the safety of the entire system?
Zeng Jice :
From the principle of data exchange or privacy computing, security verification requires proof of various algorithms, some certificates, and so on. From the system level, for example, the system has undergone security certification, 3A, 4A security certification, including the identity of the partner, authorization, account number, and audit. In addition, there is another big killer feature, because WeBank is a set of things based entirely on open source. Our entire family of open source products, including online reasoning and one-click traceability modeling will be launched soon, and will also follow the blockchain Technology integration.
We not only want to establish some standards in academia, we also try to establish some standards in industry through open source methods. The previous points just say that we have done some measures in the front, and we will continue to provide auditable after the event, including we will provide some network capture methods, as well as logs. We will use the blockchain to perform some behaviors during the entire cooperation process. Retrospective.
How to break through the traditional data cooperation model and realize joint modeling and value sharing in which new data is available and invisible?
Wu Shanshan:
Under the traditional data cooperation model, it is difficult to separate the right to use data from the ownership. Everyone agrees that the calculation, modeling, application, and analysis of each of our data will produce better results. But usually in practice, you will find that those companies that have more precious data resources are very reluctant to share their data. One is that the party receiving the data will not pay for the same data again and again. After the control of the data owner, its use is actually difficult to be clarified and regulated. This is one of the problems. The second problem is to return to the joint modeling. In the original joint modeling process, everyone outputs some desensitized data labels. It contains some information, but the granularity is coarser, so it is far from the real modeling scene. some.
Returning to federated learning, federated learning actually proposed a relatively new technical architecture, where everyone can perform joint calculations in the case of physically dispersed but logically concentrated, and the entire circulation process is available and invisible.
Challenges encountered during the entire commercialization process of private computing
Lin Lin:
The first challenge is security certification. Security is what customers care most about and is the most difficult link and step to prove. Security certification will be divided into two parts, one is theoretical security and the other is engineering security. Theoretical security is to solve this problem by completing some industry-related certifications, and also working with customers to do security-related POCs. Engineering safety, we hope that through our products, multi-party secure computing and federated learning, such a very complex but very black-core technology, can be made transparent so that users can see the actual steps of each operation. The circulation of nodes makes them feel that this technology is relatively controllable.
The second challenge is that privacy computing is also a very new technology. It is not as mature as other products in all aspects, so its application threshold is very high. We hope to pass what is already being done or already online, including processes The automation of federated learning, drag-and-drop modeling of federated learning, and some algorithm libraries and model libraries shared by federated to solve such problems, continuously lowering the threshold for users, and promoting large-scale applications.
The third challenge is that there are many technical schools in this industry, and so far, these technical schools are not compatible with each other. We also hope that in this link, we can continue to promote interconnection, and then go Push the industry one step forward.
In which areas are federated learning and private computing suitable to shine?
Lin Lin:
At present, our multi-party secure computing and federated learning technology have some very good cases in the fields of intelligent risk control, intelligent operation, open government and intelligent marketing. We recently cooperated with Bank of Communications and China Mobile in the financing service project for small and medium-sized enterprises based on multi-party secure computing, and was recently selected as a pilot application for financial technology innovation supervision of the Shanghai headquarters of the People’s Bank of China. In this project, we use the technology of multi-party secure computing to allow banks and operators to integrate the relationship graphs of the companies of the two parties under the condition that the data is not out of the domain. They can be integrated to help financial institutions identify the company behind it. The very complex network of relationships and fraud risks help financial institutions to place loans to small and medium-sized enterprises more accurately, improve their risk control level, and improve the user’s loan experience. This project is currently the first privacy computing project to enter the central bank’s regulatory sandbox . We believe that it may be a good guide for the development and application of the entire privacy computing industry in the financial industry.
Wu Shanshan: We think that privacy computing is a general-purpose technology that can make a big difference in various industries. Let’s talk about traditional agriculture. We are cooperating with banks and enterprises in agriculture. We are now working with a digital tool for agricultural wholesale markets. Companies are cooperating.
Zeng Jice: The scenarios where federal learning is used more frequently in China are scenarios such as risk control. There are small and micro credit loans, pre- and post-loan approval. Several of Weizhong’s own fist loan products also use our current vertical federation technology. At the same time, in the horizontal federation, Weizhong has a project called anti-money laundering, which is recommended by Shenzhen as a pilot project. In addition, I also made a multifaceted thing, a bit similar to insurance or rights.
Yao Ming: Privacy computing technology was born in a special environment. The premise of its birth is that multiple parties participate in the entire data transfer. The role of the scene party involved in this process includes data provider, data application party, technical service party, and government supervision. Fang etc. The gene of insight technology is to serve the credit industry, the extension is to serve the financial industry, and the extension is to serve the government and financial industries.
Xu Min: I am personally optimistic about three industries: finance, health, including social services, including intelligence, because these three industries are tailor-made solutions on the one hand, and very personal scenarios. Let me talk about the effect of the scene application. In many cases, the effect of two data collisions is very, very unexpected. We are currently working on a project to help a bank make a pre-sale risk control model. We quote operators seven to eight. This factor, when combined with the bank factor, is very efficient because the two industries were previously different industries. Operators didn’t know that their data was so useful in the financial industry, and the bank didn’t know how much data the operators had. The bridge is to carry out more efficient cooperation on the entire data, which will definitely bring more returns and will be better for the entire society.
The development trend of private computing and federated learning
Xu Min:
2020 is the first year of privacy computing, and 2021 is the year that the application of privacy computing will be implemented . I can imagine that in the next year, there will be many scenes of a hundred flowers blooming, and many scenes will be replicated by the entire industry. The Blue Elephant Alliance will focus on the financial industry, and at the same time we will focus on the two areas of financial marketing and risk control.
Yao Ming: The market’s data circulation can be divided into four eras: the first era is the era of data moving; the second era is the era of API, which is the model of instant adjustment; the third era is the era of sandbox mode; The four eras are the era of privacy computing, that is, the first year of 2020 is the “data available and not visible” and the data is not separated from the database. It is also an upgraded version of the sandbox model. The real market is the coexistence of these four models. In my opinion, the application of privacy computing must not only develop new landing scenarios, but also upgrade and iterate the traditional business model of data circulation or shared exchange. Therefore, from the perspective of the entire market, there is both a stock market and an incremental market. From a technical point of view, whether it is data moving or API calls, sandboxing, privacy computing, etc., they all have room for their own value. I don’t think a certain technology can take the entire market. Or choose to combine or superimpose a certain technology according to the characteristics of the scene to complete the commercial landing.
In layman’s terms, no matter whether a white cat or a black cat can catch a mouse, it is a good cat. There is no omnipotent technology, only a omnipotent combination. This is my point of view.
Zeng Jice: I think there must be a hundred flowers blooming in technology, and the maturity of different privacy technologies is also endless in terms of frameworks, and the frameworks will also tend to be open and mature. Second, Weizhong currently operates a relatively large cooperative network during commercialization, and many problems have been exposed. I think standards are still very important, especially in industrial applications. If we are in the system framework With a set of application scenarios, interconnection will become simpler.
Third, because the network tends to be larger and richer, governance issues are also more prominent, including how each data user and supply side is governed by his contribution, and his security, etc., governance will also become Very important.
Fourth, supervision. In this year, my country has also issued similar laws and regulations, and this field shines brightly.
Wu Shanshan: We think that in 2021, one is the gradual implementation and gradual implementation of related laws and regulations regarding the sharing and exchange of privacy information, as the privacy business continues to explode, as the market continues to respond and express themselves. Construction, of course, together with all the leaders and colleagues present here, and the common deep cultivation of our industry, we believe that in 2021, it must be driven by these, and privacy computing will have a significant evolution. So in this evolution process, you may find a more suitable route for you, and there may be some partners who will evolve to platform and deepen their cultivation along the scene. For us, we will do more work on interconnection and work with everyone to make the technology more mature, and its application scenarios will be pushed forward.
Lin Lin: First, in 2021, the acceptance of privacy computing in the entire market will rise very well, and there will also be a large number of commercialization cases to land. The second point is that we very much agree with technological advancement. We have competition behind us, but we have a lot of exchanges, and we will work together to overcome it in small technical details, and really push this technology to a higher level.
The third point is that we believe that in 2021, at the regulatory level, our industry standards will be more clear, and our entire privacy computing track and its business logic will be clearer.
The fourth point, I very much agree with the term interconnection, because from the ultimate thinking, private computing is very likely to become a very important underlying infrastructure in the future. We hope to be able to connect privacy computing to truly enable the value of data, value and AI to flow in more places in a safe manner.