In order to discuss the application and research of privacy computing technology in the financial industry, the Open Finance Alliance held an online seminar on September 30 with the theme of "Privacy Computing Helps Financial Development". The partner & solution general manager of Impulse Online was invited Participated in this event and delivered a speech entitled "Application of Privacy Computing Technology Combining Software and Hardware in the Financial Industry", and discussed with many experts and leaders how privacy computing can help the development of the financial industry.
The "Open Finance Alliance" was jointly established by 12 financial institutions in the Yangtze River Delta region, including Shanghai Pudong Development Bank, Pacific Insurance, and Guotai Junan Securities. and other in-depth cooperation to achieve complementary advantages, win-win cooperation and coordinated development.
01
Trusted Execution Environment TEE Technical Overview
In the era of big data, issues such as data circulation and security are coming one after another. Privacy computing is a technology born to solve the security problems of data circulation. It includes three technical paths: multi-party secure computing, federated learning, and trusted execution environment (TEE). Among them, the trusted execution environment is a technology that can combine privacy security and performance advantages, fully meeting the needs of high-performance computing scenarios in production environments.
TEE is the mainstream solution for privacy computing in the world. Large foreign public clouds AWS, Google Cloud, and Azure all have mature TEE products that can be purchased directly. Most of the products are built based on the Intel SGX ecosystem. Relying on domestic Xinchuang and independent and controllable demand, domestic chip manufacturers have started to make efforts in the direction of TEE. Currently, manufacturers supporting TEE chips such as Haiguang, Kunpeng, and Phytium have provided independent and controllable TEE in mass-produced chips. accomplish.
picture
The industry's first domestic privacy computing X86/ARM/self-contained all-in-one machine created by Impulse Online uses the secure area of the chip's trusted execution environment to realize data encryption and shared computing, and its performance is 30-50 times higher than that of pure software solutions. Optimize performance, easily break through the bottleneck of privacy computing software that cannot support the efficient circulation and sharing of data of more than 100 million levels, and realize standardized delivery of products.
02
Evolution of privacy computing solutions from CPU to GPU
And this year, manufacturers of computing hardware GPUs have also begun to introduce TEE functions, that is, NVIDIA (NVIDIA) released the H100 GPU series products based on the latest Hopper architecture, which is the world's first GPU accelerator with confidential computing functions. Based on the characteristics of TEE hardware acceleration by GPU, privacy computing technology can serve the collaborative tasks of larger data volume and more dimensional AI models, so that it can be extended to the entire AI field.
Image
Impulse Online proposed the first domestic TEE direct-to-GPU solution, breaking the international monopoly of Azure+Intel+Nvidia. Supports models that process more data volumes and data dimensions, and supports processing machine learning and deep learning models, greatly improving computing efficiency and lowering the threshold for application promotion.
The heterogeneous privacy computing all-in-one product based on the Haiguang CPU series chips and the impulse data interconnection platform is currently the most leading heterogeneous computing power privacy computing solution in China. Using Haiguang chips and Haiguang's enhanced virtualization isolation capabilities, it solves AI scenarios Under the trust issues of data owners, algorithm providers, computing power providers and other parties. For industry cloud/proprietary cloud/hybrid cloud, as well as various data centers that have AI capability output requirements, they can provide external services with their accumulated AI algorithms and data capabilities to ensure the privacy and security of data submitted by customers during the service process.
picture
03
Application of Privacy Computing Solution Combining Software and Hardware in the Financial Field
Finally, Huang Miao elaborated in his speech how the AI SaaS service of Impulse Online Open Banking can solve the business pain points of AI SaaS service through the application of TEE technology, that is, in the AI scene, data owners, modeling service providers, model service providers, computing The problem of distrust between power owners. Confidential computing can be used as the basic framework. The privacy computing platform manages multiple confidential computing nodes as working nodes; OCR algorithms and data to be identified (such as contracts, invoices, etc.) are run on confidential computing nodes, and sensitive data such as customer contracts and invoices It will not be placed and leaked, and the mutual trust between the data side and the model side will be realized.