不确定性系统隐私性的建模与分析

Although big data analysis has shown great value in some fields, the current development of big data is still faced with many problems. Among others, privacy is recognized as a key issue. In the context of big data, there is an intricate relevance among data. To prevent an attacker from acquiring privacy by using a priori knowledge, Dwork proposed the concept of differential privacy. In recent years, differential privacy has been broadly concerned and has become an important research field to solve the privacy problem in the context of big data. However, it is worth noting that, in some systems such as Airavat performing MapReduce computations, the differential privacy technology is only used in some components. Clearly, preserving the privacy of components is not sufficient for the privacy protection of an entire system. In order to ensure privacy at the system level, this project intends to model and analyze the privacy for uncertain systems. Based on the classical notions of differential privacy and probabilistic bisimilarity, we introduce the concept of differential privacy of an uncertain system and investigate its properties. Moreover, we design and analyze related algorithms of preserving differential privacy of a system, and apply the theoretical results to analyze and verify some currently widely used network security protocols. The project is in favor of revealing the privacy protection mechanism of systems or dynamic data, and provides an accurate and easy-to-use model and method for preserving privacy. In this way, it will lay the security foundation for the large-scale applications of big data.

尽管大数据分析在许多领域展现出巨大价值，但目前大数据的发展仍然面临着诸多问题，隐私问题便是公认的关键问题之一。在大数据背景下，数据间存在着复杂的关联性。为防止攻击者利用先验知识获取隐私，Dwork提出了差分隐私概念。近年来，差分隐私受到广泛关注，成为有望解决数据隐私问题的一个重要研究方向。值得注意的是，在诸如执行MapReduce计算的Airavat等系统中，差分隐私技术仅仅在构件层面使用。显然，保护单个构件的隐私并不意味着保护了整个系统的隐私。为了在系统层面确保隐私性，本项目拟对不确定性系统的隐私进行建模和分析，借鉴传统差分隐私和概率互相似思想，给出系统差分隐私概念，深入探讨其性质，设计确保系统差分隐私的相关算法，并将理论成果应用于分析和验证当前被广泛使用的网络安全机制。本项目的开展有利于揭示系统或动态数据的隐私保护机制，提供准确、易用的隐私保护模型和分析方法，为大数据产业奠定安全基础。

近年来，差分隐私受到了广泛关注，成为有望解决数据隐私问题的一个重要研究方向。值得注意的是，在诸如执行MapReduce计算的Airavat等系统中，差分隐私技术仅用于构件层面。显然，保护单个构件的隐私并不意味着保护了整个系统的隐私。为此，本项目组从系统层面研究了隐私保护，建立了一套较完整的不确定性系统中的隐私保护理论框架。基于传统差分隐私和概率互相似的思想，我们从系统层面建模和分析了不确定性系统的隐私性，给出了不确定性系统差分隐私的概念，深入研究了系统内部差分隐私的性质，揭示了其与传统差分隐私和概率互相似的关系，模块化分析了不确定性系统差分隐私的变化规律，并将理论研究成果应用于隐私通信协议——Crowds协议和机器学习中训练集的成员隐私保护。本项目的顺利开展，丰富和发展了隐私保护的理论与方法，为系统地保护用户数据的隐私性提供了技术支持。