隐私计算关键技术与创新

Key technology and innovation of privacy preserving computing

数据已经成为数字经济中重要的生产要素和战略资源。在实际应用中,数据分散在各个机构,由于隐私保护等原因,这些数据很难整合到一起,形成数据孤岛,从而阻碍大数据与人工智能产业的发展。隐私计算可以破解数据孤岛难题,在保障用户隐私的同时赋能多方数据协同应用,助力释放数据融合价值。对隐私计算的关键技术与创新成果,包括隐私集合求交技术、斜向联邦学习、异步并行计算、消息压缩机制、单向通信连接方案、可信执行环境、联合数据分析等进行了研究和分析,并对这些技术与创新在Angel PowerFL通用隐私计算平台中的应用进行了介绍。

Big Data has become one of the factors of production and a strategic resource in the digital economy.However,data are usually scattered in different organizations and cannot be integrated in a centralized manner due to privacy concerns in practice,hindering the performance of many real-world big data applications.In order to tackle this problem,privacy preserving computing has been developed to enable the application of data from different parties for federated learning and analysis with privacy guarantees.In this paper,we provide a survey of the key techniques and advances in privacy preserving computing,including private set intersection,diagonal federated learning,asynchronous parallel computation,message compression protocols,unidirectional connections,trusted execution environments,and federated data analysis.Finally,we introduce the applications and techniques in Angel PowerFL,which is a general and industry-grade privacy preserving computing platform.