人工智能模型数据泄露的攻击与防御研究综述

人工智能和深度学习算法正在高速发展,这些新兴技术在音视频识别、自然语言处理等领域已经得到了广泛应用。然而,近年来研究者发现,当前主流的人工智能模型中存在着诸多安全隐患,并且这些隐患会限制人工智能技术的进一步发展。因此,研究了人工智能模型中的数据安全与隐私保护问题。对于数据与隐私泄露问题,主要研究了基于模型输出的数据泄露问题和基于模型更新的数据泄露问题。在基于模型输出的数据泄露问题中,主要探讨了模型窃取攻击、模型逆向攻击、成员推断攻击的原理和研究现状;在基于模型更新的数据泄露问题中,探讨了在分布式训练过程中,攻击者如何窃取隐私数据的相关研究。对于数据与隐私保护问题,主要研究了常用的3类防御方法,即模型结构防御,信息混淆防御,查询控制防御。综上,围绕人工智能深度学习模型的数据安全与隐私保护领域中最前沿的研究成果,探讨了人工智能深度学习模型的数据窃取和防御技术的理论基础、重要成果以及相关应用。     

个性化时空数据隐私保护

智能移动终端的普及导致收集的时空数据中个人位置隐私、签到数据隐私、轨迹隐私等敏感信息容易泄露,且当前研究分别针对上述隐私泄露单独提出保护技术,而没有面向用户给出防止上述隐私泄露的个性化时空数据隐私保护方法.针对这个问题,提出一种面向时空数据的个性化隐私保护模型(p,q,ε)-匿名和基于该模型的个性化时空数据隐私保护(P...     

隐私数据验证场景下的隐私保护研究

隐私数据验证场景是信息验证服务下的一类特殊场景,其实用性要求数据在第三方数据库进行存储、发布且有能力处理任意形式声明的验证,其安全性要求数据在存储、更新与证明期间提供有效的隐私保护手段。目前该场景下的隐私保护研究尚且处于空白阶段,因此本文引入可证明数据加密策略的概念,以满足隐私数据验证场景下的实用性与安全性需求。本文主要有三个贡献：（1）对可证明数据加密策略进行讨论并给出形式化定义;（2）基于非交互零知识证明构造出首个可证明数据加密方案,并同时支持高效的数据更新操作;（3）基于承诺方案、非交互零知识证明与全同态加密,提出可证明数据加密策略的两种通用构造框架并给予相关性质证明。     

对混合属性数据表可行的差分隐私保护方法

为加强隐私保护和提高数据可用性,提出一种可对混合属性数据表执行差分隐私的数据保护方法。该方法首先采用ICMD（insensitive clustering for mixed data）聚类算法对数据集进行聚类匿名,然后在此基础上进行
-差分隐私保护。ICMD聚类算法对数据表中的分类属性和数值属性采用不同方法计算距离和质心,并引入全序函数以满足执行差分隐私的要求。通过聚类,实现了将查询敏感度由单条数据向组数据的分化,降低了信息损失和信息披露的风险。最后实验结果表明了该方法的有效性。     

面向无线传感器网络的数据完整性和隐私保护融合算法

无线传感器网络(WSNs)作为物联网的重要组成部分,在实际应用中,希望在得到精确数据融合结果的同时,又能保护数据信息的隐私性和完整性。为此,提出一种新的数据融合完整性保护算法,在增添私有种子对节点采集数据进行隐私保护的基础上,利用复数的虚部数据与采集到的真实数据呈非线性关系,有效地实现信息完整性的鉴别。性能分析和仿真结果表明:该算法可以在较低数据通信开销与计算开销的前提下,应对恶意节点的各种攻击,提供更有效更可靠的数据完整性保护。     

基于差分隐私的数据匿名化隐私保护方法

在保护数据隐私的匿名技术中,为解决匿名安全性不足的问题,即匿名过程中因计算等价类质心遭受同质性和背景知识攻击造成的隐私泄漏,提出了一种基于差分隐私的数据匿名化隐私保护方法,构建了基于差分隐私的数据匿名化隐私保护模型;在利用微聚集MDAV算法划分相似等价类并在匿名属性过程中引入SuLQ框架设计得到ε-MDAV算法,同时选用Laplace实现机制合理控制隐私保护预算。通过对比不同隐私保护预算下可用性和安全性的变化,验证了该方法可以在保证数据高可用性的前提下有效地提升数据的安全性能。     

Auditing images collected by sensors in ambient intelligence systems with privacy and high efficiency

The Journal of Supercomputing - To determine whether sensors should transmit collected images back to the data center, an auditing protocol is desired to check these images before transmission....     

Discriminant component analysis for privacy protection and visualization of big data

Multimedia Tools and Applications - Big data has many divergent types of sources, from physical (sensor/IoT) to social and cyber (web) types, rendering it messy and, imprecise, and incomplete. Due...     

Singular value decomposition based data distortion strategy for privacy protection

Privacy-preserving is a major concern in the application of data mining techniques to datasets containing personal, sensitive, or confidential information. Data distortion is a critical component to preserve privacy in security-related data mining applications, such as in data mining-based terrorist analysis systems. We propose a sparsified Singular Value Decomposition (SVD) method for data distortion. We also put forth a few metrics to measure the difference between the distorted dataset and the original dataset and the degree of the privacy protection. Our experimental results using synthetic and real world datasets show that the sparsified SVD method works well in preserving privacy as well as maintaining utility of the datasets. Shuting Xu received her PhD in Computer Science from the University of Kentucky in 2005. Dr. Xu is presently an Assistant Professor in the Department of Computer Information Systems at the Virginia State University. Her research interests include data mining and information retrieval, database systems, parallel, and distributed computing. Jun Zhang received a PhD from The George Washington University in 1997. He is an Associate Professor of Computer Science and Director of the Laboratory for High Performance Scientific Computing & Computer Simulation and Laboratory for Computational Medical Imaging & Data Analysis at the University of Kentucky. His research interests include computational neuroinformatics, data miningand information retrieval, large scale parallel and scientific computing, numerical simulation, iterative and preconditioning techniques for large scale matrix computation. Dr. Zhang is associate editor and on the editorial boards of four international journals in computer simulation andcomputational mathematics, and is on the program committees of a few international conferences. His research work has been funded by the U.S. National Science Foundation and the Department of Energy. He is recipient of the U.S. National Science Foundation CAREER Award and several other awards. Dianwei Han received an M.E. degree from Beijing Institute of Technology, Beijing, China, in 1995. From 1995to 1998, he worked in a Hitachi company(BHH) in Beijing, China. He received an MS degree from Lamar University, USA, in 2003. He is currently a PhD student in the Department of Computer Science, University of Kentucky, USA. His research interests include data mining and information retrieval, computational medical imaging analysis, and artificial intelligence. Jie Wang received the masters degree in Industrial Automation from Beijing University of Chemical Technology in 1996. She is currently a PhD student and a member of the Laboratory for High Performance Computing and Computer Simulation in the Department of Computer Science at the University of Kentucky, USA. Her research interests include data mining and knowledge discovery, information filtering and retrieval, inter-organizational collaboration mechanism, and intelligent e-Technology.     

具有隐私保护的边缘计算高效数据卸载方法

针对用户终端数据卸载过程中难以同时实现高隐私安全和低时间消耗的目标,提出了一种具有隐私保护的边缘计算高效数据卸载方法.首先,利用时间计算模型和隐私熵值分别将用户终端时间消耗和数据隐私安全程度进行量化,并建立一个多目标优化问题模型;其次,利用改进强度帕累托进化算法对时间消耗和隐私熵值进行联合优化;最后,利用基于熵权法的多属性决策方法选取最优的时间消耗和隐私熵组合策略.在多终端用户多计算任务的边缘计算下展开实验研究和对比分析,结果表明,该方法在降低传输时间的基础上还增强了数据卸载传输的安全性.     

Protecting business intelligence and customer privacy while outsourcing data mining tasks

Nowadays data mining plays an important role in decision making. Since many organizations do not possess the in-house expertise of data mining, it is beneficial to outsource data mining tasks to external service providers. However, most organizations hesitate to do so due to the concern of loss of business intelligence and customer privacy. In this paper, we present a Bloom filter based solution to enable organizations to outsource their tasks of mining association rules, at the same time, protect their business intelligence and customer privacy. Our approach can achieve high precision in data mining by trading-off the storage requirement. This research was supported by the USA National Science Foundation Grants CCR-0310974 and IIS-0546027.

---

Comparative legal study on privacy and personal data protection for robots equipped with artificial intelligence: looking at functional and technological aspects

AI & SOCIETY - This paper undertakes a comparative legal study to analyze the challenges of privacy and personal data protection posed by Artificial Intelligence (“AI”) embedded in...     

一种基于区块链的医疗数据隐私保护方法

为解决医疗数据的泄露或恶意被窜改以及医疗纠纷问题,提出一种基于区块链的医疗数据隐私保护方法。利用哈希算法加密患者的身份信息,治疗结果通过AES（advanced encryption standard）算法加密,而AES的密钥使用ECC（ellipse curve ctyptography）算法加密,所有的加密密钥、治疗结果、患者身份信息存储到联盟链上。采用群签名技术追溯签名医院,群管理员可以解密医疗数据,将其作为重要依据协助第三方解决医疗纠纷。效率分析表明,在安全性相同情况下,该方法较的加/解密效率比对比方案分别提高了14%和46%,同时分析了群签名各类算法的时间开销。通过与同类方法对比,该方法既可实现患者身份、医疗数据的分类隐私保护,又可保证交易存储开销是合理的,在医疗数据隐私保护领域具有一定的应用价值。     

基于迁移学习的多源数据隐私保护方法研究

隐私保护的多源数据分析是大数据分析的研究热点,在多方隐私数据中学习分类器具有重要应用。提出两阶段的隐私保护分析器模型,首先在本地使用具有隐私保护性的PATE-T模型对隐私数据训练分类器;然后集合多方分类器,使用迁移学习将集合知识迁移到全局分类器,建立一个准确的、具有差分隐私的全局分类器。该全局分类器无需访问任何一方隐私数据。实验结果表明,全局分类器不仅能够很好地诠释各个本地分类器,而且还可以保护各方隐私训练数据的细节。     

大数据环境下隐私保护技术研究

大数据环境下安全问题面临着新的形势和新的矛盾,其中隐私保护是关键的问题之一,如何在大数据环境下兼顾共享和隐私保护是不得不考虑的问题。文章分析了大数据环境下隐私保护问题面临的新困难,指出隐私保护问题实际上就是数据共享问题,初次共享可以通过授权控制,但是数据的过度共享无法单单通过授权来控制。研究新的使用控制模型,提出通过授权、职责和条件来控制非授权的数据过度共享。在此基础上提出了职责后使用的控制模型,通过职责操作和条件约束来控制信息数据的过度共享,并给出了该模型的形式化描述。     

基于SOM网络的差分隐私保护研究

差分隐私模型具有强大的隐私保护能力,但是也存在数据效用低等问题。为提高数据可用性并保护数据隐私,提出一种基于SOM网络的差分隐私算法(SOMDP)。首先利用SOM网络模型对数据进行聚类操作;其次,对每个划分好的聚类添加满足差分隐私的拉普拉斯噪声;最后,理论分析算法的可行性,并在真实数据集上评估SOMDP算法性能、算法的数据可用性和隐私性能。实验结果表明,SOMDP在达到差分隐私要求的前提下,可较大程度地提高差分隐私数据发布的效用。     

Video security for ambient intelligence

Moving toward the implementation of the intelligent building idea in the framework of ambient intelligence, a video security application for people detection, tracking, and counting in indoor environments is presented in this paper. In addition to security purposes, the system may be employed to estimate the number of accesses in public buildings, as well as the preferred followed routes. Computer vision techniques are used to analyze and process video streams acquired from multiple video cameras. Image segmentation is performed to detect moving regions and to calculate the number of people in the scene. Testing was performed on indoor video sequences with different illumination conditions.