If we talk about the privacy computing track, what are the recent hot keywords?
"Open source" is one of them.
With the acceleration of the data element market cultivation, privacy computing is a key technical solution for the safe circulation of data, how to speed up its technology development and commercialization has become a market concern.
From the first year of commercialization in 2020 to this year, in the in-depth running-in and matching of privacy computing technology service providers and B-end customers, the market has become more and more consensus that in the "commercial blueprint" of privacy computing, high-performance computing power, open source Ecology, software and hardware engineering optimization, and scene adaptation have become "standard items" and are indispensable. "Open source" is regarded as an important path and technical means for privacy computing to improve performance and scale up.
In October last year, the central bank and other departments jointly issued the "Opinions on Regulating the Application and Development of Open Source Technology in the Financial Industry", emphasizing that "encourage open source technology providers, accelerate the improvement of technological innovation capabilities, earnestly master the core code of open source technology, form independent intellectual property rights, and consolidate Industry Support Capabilities”.
The computing power think tank found that from this year to August, Ant Group has successively announced the open source privacy computing framework "Yiyu", Jiuzhang Yunji released the YLearn causal learning open source project, Yuanyu Technology launched the privacy computing open source platform Primihub, Yifang Jianshu Announcing the open-source federated learning of Wings and secure calculation of Wings, it can be seen with the naked eye that open source is gradually "popular", and the open source team is no longer exclusive to big companies such as Ali, Byte, and Baidu. Forces are also being added one after another.
At the same time, in May of this year, the first international independent and controllable privacy computing open source community in China, the Open Islands open source community, jointly initiated by nearly 50 units of industry, academia and research institutes, was also officially established.
"Open source swallows everything", this is the motto of geeks, embracing open source has become the mainstream of the global basic software industry. In the past 25 years, open source has driven most of the technological innovations. From applications, to every website and platform you browse, to the collaborative interaction between everything in the Internet of Things era, it can be said that more than 90% of the code in the world has open source behind it.
For privacy computing, which is still in its infancy, the open source revolution has just begun.
1Privacy
Computing Erlian Q:
Why open source? Why now?
As the infrastructure of data circulation, the necessity of open source privacy computing lies not only in realizing the general requirements of optimization and iteration of the technology itself, but also based on the particularity of the circulation of its service data elements.
"If privacy computing and federated learning technology are only in the hands of a few oligarchs, we still cannot get real data circulation, nor can we get real digital economic development. Therefore, the threshold must be lowered. One of the effective means is open source. Everyone can use this kind of technology, and everyone can contribute to this kind of technology", said Yang Qiang, a chair professor of the Department of Computer and Engineering of the Hong Kong University of Science and Technology and chairman of the FATE Open Source Community Technical Steering Committee. Executive Chairman of the Open Islands open source community.
From a goal-oriented point of view, the marginal benefit of data use is increasing. Only by realizing a widely circulated data element market can we create and release greater data value. This means that we must open up the closed loop of infrastructure and eliminate technology islands. If giants They are all out of commercial profit, and the implementation of closed technology monopoly is not conducive to the sustainable long-term development of privacy computing. For privacy computing, a technical attribute that "acts on and inhabits" data circulation scenarios, openness , Inclusiveness is its proper meaning. A practitioner in the privacy computing industry told the computing power think tank.
Yang Qiang also expressed the same point of view. He believes that on the commercial roadmap of privacy computing and federated learning, security, efficiency, effectiveness, and inclusiveness are the latitude lines, and the technology iteration and scene popularization dominated by open source ecology are the longitude lines. Open source promotes The "inclusiveness" of privacy computing is symbiotic with value. Take FATE, an open source community for federated learning, as an example. The open source of FATE has opened up the wave of open source privacy computing technology in China, effectively lowering the technical threshold of "federated learning". Or referring to open source projects, the FATE open source community has accelerated the coverage and popularization of Federated Learning from "big factories" to small and micro B-end enterprises, and at the same time allowed the federated learning industry ecology and participants to move from "individual combat" to ecologicalization.
In addition, another obvious reason is that "from the supply side of technology development, using existing resources, there is no need to repeat development, reinvent the wheel, stand on the basis of existing technology, and grasp the already tempered and verified ecosystem. And scenes, and then delve into adding your own innovations, and it will not cause waste of technical resources.
From the demand side of banking, finance, medical institutions, etc., there are inherent barriers to interconnection and interoperability of privacy computing products with different technical routes. A very important issue is that privacy computing is driven by algorithms, and its backdoor risk of "algorithm black box and data black box" also arises. Although privacy computing manufacturers have always promised to be "safe, reliable and reliable" and will not steal and retain data, but How can we really win the trust of others and prove our innocence? Wang Lei, general manager of Ant Group's Privacy and Intelligent Computing Department and head of the "Argot" framework, also said: "From a technical perspective, if others can't see our code, they can't confirm the security of the product, so what's the point of trust? The way of sharing attracts more excellent developers to join in, so as to unite technology and work together to reduce the technical threshold of privacy computing developers and users."
It can be observed that in recent years, whether it is regulatory compliance, personal information protection, or business risk control, the interpretability and security requirements for algorithms and models have become higher and higher. For example, in March 2021, the central bank issued and The implementation of the "Artificial Intelligence Algorithm Financial Application Evaluation Specification" requires that the application of AI algorithms must meet security and explainability; at the end of 2021, the "Internet Information Service Algorithm Recommendation Management Regulations" jointly issued by four ministries and commissions, among them In terms of user rights protection, In particular, it is stipulated that the algorithm recommendation service provider shall inform users of its provision of algorithm recommendation service in a prominent manner, and publicize the basic principles, purpose and main operating mechanism of the algorithm recommendation service in an appropriate manner.
"Explainability" and "zero trust" should become the genes of technology, and privacy computing is no exception. Open source can help users understand its technical logic through the public verifiability of full code, and promote the transparency of technology, so that it can be self-evident." , the aforementioned practitioner in the privacy computing industry continued.
It is worth noting that in the past few years, open source has not become popular in the privacy computing circle, but now, open source has become popular and the voice is loud.
"The reason why the open source and openness of privacy computing is getting more and more attention is because it is just at the right time. First of all, it is in line with the trend of a unified national market. At the beginning, we emphasized more about which technology is used for privacy computing, rather than specific I am concerned about the purpose to be achieved, so it may be a bit off track. For example, some manufacturers especially emphasize that privacy computing must use multi-party computing to be safe, and it is not safe to use other technologies. Therefore, in terms of technology selection, A financial institution uses a B uses another type of technology, C may be a big data company, and the third type of technology used has different standards, so that when everyone wants to communicate with each other, they find it difficult to communicate with each other. Therefore, it is very timely to propose a unified market; secondly, to a certain extent, open source is also a sign of the gradual maturity of privacy computing. More and more companies choose open source. First, they believe in their own products and technical strength. The large-scale application and innovation of privacy computing provides more efficient solutions. Based on open source collaboration, the participation of more roles such as users and ecological partners enables the technology to accept more dimensional inspections, and can also establish a more agile and comprehensive response The mechanism can respond to security risks at any time, which greatly improves the security and iteration efficiency of software algorithms." Yang Qiang said.
2
Technical horse racing, what high-quality open source projects are there?
Open source has become a "trend", and high-quality players gather.
According to incomplete statistics from the computing power think tank, many large companies and entrepreneurial teams at home and abroad have been actively open source in recent years.
Table 1: Main open source frameworks/platforms for privacy computing
(Data statistics: Institute of Information and Communications Technology, Computing Power Think Tank)
The current privacy computing open source projects can be roughly divided into protocol framework open source and product open source. Most of the protocol framework open source is aimed at a certain technology, such as mp-spdz and OpenCheetah in the MPC field, focusing on security and performance improvement. In addition, there is also open source for product platforms, which makes it easier to form an ecology. Generally speaking, excellent low-level open source protocols can be embedded into the platform and widely used, while most of the open source projects of privacy computing products are still in their infancy, only the code is open but the community building is not complete. The above table shows the main open source platforms or protocol frameworks at home and abroad. It can be seen that in the past three years, more and more enterprises have joined the open source team of privacy computing, including underlying technology protocols and enterprise platform projects.
In the face of the many open source parties in the market, which indicators do developers and users pay more attention to? A technical director of a privacy computing company revealed that among the various open source frameworks for privacy computing, the open source frameworks of federated learning and multi-party secure computing are the majority. These two technical paths are relatively mature and gradually becoming mainstream. When cooperating with some large commercial banks, they usually consider self-research on a mature framework and start from joint development.
Wang Lei, general manager of Ant Group's Privacy Intelligent Computing Technology Department, also pointed out that banks mainly focus on the ease of use and compliance of technology when bidding and co-construction. If a framework has a high threshold for use, it will be difficult to use it. However, the industry is still in the groping stage in this regard.
Objectively speaking, software ecological construction is more difficult than the development of the software itself. If privacy computing wants to achieve industrial-scale applications, it needs to do a lot of things beyond privacy computing. Ecological construction is a key step, and open source and openness can enhance The viscosity between all walks of life in the ecology.
Zhang Lintao, chief scientist of Yifang Jianshu, also said: With the development of technology, more and more players in the industry have already possessed considerable technical strength. If they want to further widen the competition gap, they must have a deeper insight into the industry instead of Again, purely technical issues. The history of open source in new technology fields such as artificial intelligence has provided a reference for privacy computing, and it has become more difficult to obtain absolute technical advantages. The emergence of open source frameworks such as TensorFlow and PyTorch has turned to attract more people into AI on top of technological competition. The track promotes the overall development of AI.
It can be seen that "open source" is starting the next journey of privacy computing technology horse racing. From competing technology to re-ecology, the values and position of the entire track are beginning to tilt towards "more inclusiveness, scalability and connectivity". Adults Daji, working together to build an open source ecological community and data element market is the long-term development path.
References
China Business News "Privacy computing open source innovation data market is expected to speed up"
Economic Observer "Yang Qiang: Why should privacy computing be open source?" "
Leifeng.com leiphone" Ant "cryptic language" open source, crossing the "boundary gap" of privacy computing"
SegmentFault thinks "We don't know anything about the power of "open source", but we look forward to it"
Privacy Computing Alliance "Release | 2022 Privacy Computing Top Ten Observations