Joint Computing and Privacy Designs for Cloud Data Services
MetadataShow full item record
Cloud computing provides a cost-effective and scalable manner to enable powerful and ubiquitous applications. The fast-growing amount of user contributed data is collected using cloud platforms. While providing the tremendous advantages, the processing of the enormous information sources also poses a considerable threat to individual users’ privacy. The existing practice of cloud computing, however, loses control of the privacy and thus leads to increasing public criticisms and legislation pressures. In this thesis, we focus on exploring the challenges and solutions to tackle practical problems for computing and data privacy in the cloud environment. In particular, we discuss the problem of cloud computing by focusing on both functionality and privacy using a case study of secure image data feature detections. Moreover, we also study the data privacy preservation in cloud using a case study of heavy hitter estimation with rigorous utility and privacy guarantee. In the light of computing in the cloud, we first focus on the case of cloud-assisted secure image data computation, which is motivated by the enormous and ubiquitous image data generated by end-users every day. In our works, both the image global feature, e.g., color layout descriptor, and local feature, e.g., scalar invariant feature transform, detection computation tasks are studied as target cloud-assisted applications. Specifically, the approaches that enable the third party to perform image feature detection algorithms over encrypted image data are developed. For image global features, to tackle the challenge that lies in the diversity of various features, a general solution based on a modified ring learning with error somewhat homomorphic encryption scheme is proposed. For image local features, to tackle the challenge that lies in the complex of feature detection algorithms, customized solutions for prevalent local features using light-weighted cryptography tools are proposed. Moreover, their performances are evaluated on real-world applications. Different from cloud computing privacy, the privacy preservation of cloud data focuses on learning meaningful conclusion without compromising individual data privacy. To handle the problem, the notion of local differential privacy is proposed to quantify the privacy guarantee of the data utilization mechanism. In this privacy paradigm, each user perturbs her data locally before sending the noisy data to a data collector. The latter then analyzes the data to obtain useful statistics. In this thesis, we propose a two-phase mechanism for obtaining accurate heavy hitters with local differential privacy. The strategy is to first gather a candidate set of heavy hitters using a portion of the privacy budget, and focus the remaining budget on refining the candidate set in a second phase, which is much more efficient budget-wise than obtaining the heavy hitters directly from the whole dataset. The proposed solution provides both sound theoretical guarantee and experimental performance that significantly improves over existing methods.