Intel Ai Baijia Innovation Plan 丨 Improvement Online: Let data flow safer

With the rapid development of the Internet, more and more data, privacy, property, etc. are stored in the network, and credible identity security has become quite important. Simple password scheme is increasingly difficult to take responsibility for ensuring user identity. In the era of data security issues, how to ensure the security of the network and data transmission, become an urgent problem to be solved.

In the beginning of the assessment, the pain points of the data circulation safety problem were regarded. From the process of circulating the circulation of enterprise data, the original data did not leave local, and the number of secure and convenient data flow and ecology can be performed with multiple external agencies. cooperation. The impulse is committed to solving problems encountered in data security circulation and is a navigator of data circulation.

With leading privacy computing technology and innovative industry applications, the impulse online from hundreds of AI companies, successfully selected the "Intel Ai Baijia Innovation Incentive Plan" Fifth. This program is based on Intel’s comprehensive AI product portfolio and unique advantages of hardware and software synergies, joint industrial ecological partners, using 3 years, accelerating at least 100 China Excellent AI innovation enterprises.

Improvement Online is a technological innovation company focused on data intelligent interconnection solutions, which is committed to promoting data production factors in social interconnection, building trusted, safe, privacy, fairness, and efficient "data internet". At the moment of data security, the impulse online has successfully provided data privacy computing services to customers in government affairs, finance, operators, and Internet and other industries. At the same time, it also strives to achieve technical landing.

The operator industry, the impulse online is a trusted data exchange platform for an operator group, establishing a data circulation platform between the group’s branches, and has implemented practices in several provincial branches, realizing customer information inside the branch. Other systems or sharing and collaboration of marketing units. The "one thing to" "one thing" sharing mode originally revealed, optimized to the current data task approval system.

The financial industry, the impulse is online for a large commercial bank to build a privacy calculation combined modeling platform. Enter multiple external data sources by privacy calculations, integrate multi-party bills, tax, business, justice, etc., machine learning on the bank Wind control model training, which helps banks improve the accuracy of wind-control model through multi-dimensional data, while reducing the bank itself uses multi-data generated user privacy information protection compliance risks.

The government industry, the impulse online is a government affairs data operation platform to build a privacy computational government affairs data sharing platform, support the government – political sharing, government – enterprises and data open, establish a data circulation platform between government and government, government and enterprises. The platform has been practiced in several government committees and enterprises, achieving the sharing of customer information and government system between enterprises, and ensuring that the sensitive data of government affairs is neither in the premise of satisfying enterprise testing results and data use. Leaks can be shared.

In fact, the design of the data circulation scheme is not easy. It is necessary to consider privacy, security, credibility, and intelligent four aspects. Privacy calculations can solve this problem, ensure that data is "not visible" in the cooperation process, that is, all parties only The results of data calculations can be obtained without obtaining raw data, and this security feature can be argued by technical means.

Specifically, privacy computing includes a variety of different technical routes, where the mainstream is based on chip-based trusted enforcement environments, censilation-based security multi-party computing (MPC), and federated federation from artificial intelligence technology Machine study.

The trusted execution environment Tee (Trusted Execution Environment) is a secure area built by a hardwareware method on a computing platform to ensure that the code and data loaded in the secure area are protected in confidentiality and integrity. The trusted execution environment goal is to ensure that a task is executed in accordance with expected, ensuring the confidentiality, integrity of the initial state, and the confidentiality and integrity of the runtime status. At present, Intel SGX is the highest thermality and the most complete TEE scheme.

Safety multi-party calculation

Federal machine learning is the expansion of machine learning technology, which can complete a multi-party commonly involved model training in the case of all parties, and effectively help multiple agencies to solve models in meeting privacy data protection. The problem of insufficient training data. Federal machine learning technology can solve complex algorithm modeling problems, but there is a bottleneck in performance.

To construct a trusted data circulation scenario based on TEE and block chains, the impulse is proposed online, which includes data demand and data owners and both part of the role. If two parts of the character are combined, the system needs to be divided into two levels, namely the control level and execution level, and the league link provides data collaboration management services. The proposal of this program marks that the impulse online is a new step in the field of data circulation, which will force data circulation technology to land, and help the development of data privacy computing industry.

Specifically, traditional federal learning frameworks will encounter two technical problems when they actually land: First, federal learning is based on a large number of complex cryptographic algorithms to ensure the security of exact process data, compared to the plaintest machine learning framework Performance loss is above one to two orders of magnitude, which cannot meet the analysis accuracy requirements in a high model timeliness requirement. Second, the federal learning framework introduces the centralized parameter server, all parties need to summarize the calculation results of each round to the parameter server, the parameter server is the key to the trust of the whole system. The parameter server is attacked or runs certain unsafe protocols, which will result in the disclosure of sensitive data and the damage of the global model, and the trust root is urgently needed.

Intel SGX2 As the most mature TEE solution, Intel SGX2 not only provides a trusted measurement root, but also provides a high-performance secretly computing environment. Provide a pragmatic technical option for federal learning to solve the above problems.

The impulse online in this topic, realizes and tested the TEE-based federal learning framework. The overall architecture uses Intel Ice Lake / SGX2 as a resource base to deploy the impulse federal learning platform on the hardware node. The impulse online will optimize the operator based on Intel Ice Lake / SGX2 to the calculation of only 30% computational overhead. At the same time, Intel SGX nodes ensure security and privacy in data collaboration. SGX2 excellent performance performance and cross-node’s masonry calculation domain provide new ideas for federal learning security sections. Imposion Online Combined with TEE and Differential Privacy Model Protecting Process Data from Domains, the training performance increases the training performance to a quantity level compared to traditional codenic methods based on homogeneous encryption.

This issue has reached the expected design effect, and it has grown out the engineering method of complex distributed systems to TEE operation. Finally, the finalization of TEE’s federal learning performance will be a project research direction worth exploring. The subsequent impulse online will work to create a high-efficiency privacy computational solution in optimized privacy computational technical performance and Intel’s continuous cooperation.

Due to the rapid development of information technology, the continuous evolution of personalized services, the frequent cross-border, cross-system, and inter-biochemical interactions of personal information data, exacerbated privacy information intentionally or unintentionally in different information systems, The privacy information protection of privacy information, the problem of tracing of privacy violations is increasingly serious, and the existing privacy protection program has not provided systematic protection.

With the "Data Safety Law" in 2021, the data industry regulation changes, requiring data sharing under the premise of protecting data security and personal privacy, and data "can be invisible" has become an indispensable new infrastructure.

In the future, the impulse online will continue to build a professional data circulation solution, strictly protect personal and enterprise data privacy information. Under the power of Intel’s powerful confidential computing hardware, it will accelerate the problem of technical problems to help more use landing, promote the large-scale landing of data elements, and play the true value of data elements.