Frontier Technology | Privacy Computing: Putting an "Invisibility Cloak" on Data

By Alibaba DAMO Academy

Currently, a new round of global scientific and technological revolution and industrial transformation is accelerating. The strategic position of digitalization in the post-epidemic era is becoming increasingly prominent. As a key resource, the huge value contained in data is being released, but the security risk of data is becoming increasingly prominent. Learning how to effectively protect the users' data privacy is an urgent problem in the development of science and technology.

Introduction

Data security protection and circulation are two major issues in the digital age, but the solution is privacy computing.

Privacy computing is not a new thing. As early as 1979, the forerunners of computer science began to explore the technical problems of secret sharing. Only a few years later, Academician Yao Qizhi, who returned to work in China later, proposed the idea of Secure Multi-Party Computation and Garbled Circuit.

However, the unique development of privacy computing in the 21st century is inseparable from that time. In recent years, the rapid development of the Internet has made privacy issues increasingly prominent, and the balance between personal privacy and network development has changed from a social issue to a scientific issue. Secure multi-party computation (cryptography), federated learning, and trusted computing environments have developed into several major areas of privacy computing.

In the past, privacy computing can only be applied in scenarios with a small amount of data due to performance bottlenecks, insufficient technical trust, and inconsistent standards. With the integration and development of technologies (such as dedicated chips, encryption algorithms, whitebox implementation, and data trusts), private computing is expected to span large amounts of data and expand data sources to the entire domain, stimulating new productivity in the digital era.

It is expected that privacy computation will make breakthroughs in performance and interpretability over the next three years. Data trust institutions may provide data sharing services based on privacy computing.

Interpretation of Application Trends

In the digital economy era, data has become the core production factor. At the same time, factors (such as data rights confirmation, data regulations, privacy protection awareness, and data security assurance) have become issues that must be faced in data sharing and value mining across organizations.

Privacy computing integrates disciplines, such as cryptography, artificial intelligence, and chip design. It uses secure multi-party computation, differential privacy, and trusted computing as representative technologies. It can implement computing and analysis while ensuring that data privacy is not leaked. It provides a feasible model for cross-organization data sharing. However, due to performance bottlenecks, insufficient technical trust, and inconsistent standards, privacy computing can only be applied in scenarios with a small amount of data.

Privacy computing will usher in three breakthroughs, enabling large-scale applications. First, the performance and efficiency of privacy computing will be improved by leaps and bounds. The improvement includes homomorphic encryption algorithm breakthroughs, reduced hash rate requirements for encryption and decryption, hardware and software acceleration chips, performance optimization for secure multi-party computation and federated learning scenarios, and a third-party trusted execution environment (TEE). The second breakthrough is the whitebox implementation of private computing. It improves the interpretability of the technology, strengthens trust, and reduces the integration barriers across technologies and models through open integration capabilities. The third breakthrough is the emergence of data trusts. As a trusted third-party, it provides technology and operations and accelerates data sharing between organizations.

The technological breakthrough of privacy computing will push data computing from the private domain to the global domain. The accuracy and depth of analysis will also increase with the available data volume. The effect is more significant in some fields that rely heavily on data volume, such as business analysis, risk control, academic research, artificial intelligence, and precision marketing. In addition, when the privacy computation is mature, it is expected to become the standard for data sharing, and the risk of data circulation will be reduced significantly. The responsibility boundary between data owners and data custodians will be clearer, and the degree of security will be more measurable.

In addition to technology, the greatest uncertainty in privacy computing comes from operating models and compliance standards. The operation models have not formed a complete system yet, in which data providers have sufficient incentives to share data when ensuring data quality so that data users are willing to pay fees. As far as compliance standards are concerned, the compliance red line of privacy computing is not clear, which makes the development of technology uncertain. Technology and standards need to promote each other in the development process.

Alibaba DAMO Academy predicts that within three years, privacy computation will have breakthroughs in performance and interpretability, and data trusts will begin to appear to provide data sharing services based on privacy computing. Global privacy computing will change the existing data circulation methods, and new businesses will also be born based on global data over the next five to ten years, improving the production efficiency of the whole society with data as the core.

Community

Frontier Technology | Privacy Computing: Putting an "Invisibility Cloak" on Data

Introduction

Interpretation of Application Trends

Read previous post:

Read next post:

Alibaba Cloud Community

You may also like

Comments

Alibaba Cloud Community

Related Products

Quick Starts

Data Security on the Cloud Solution

Function Compute

Data Lake Formation