Abstract

The privacy preserving issues have received significant attentions in various domains. Various models and techniques have been proposed to achieve optimal privacy with minimal costs. However, side-channel leakages (such as, publicly-known algorithms of data publishing, observable traffic information in web application, fine-grained readings in smart metering) further complicate the
process of privacy preservation. In this thesis, we make the first effort on investigating a general framework to model side-channel attacks across different domains and applying the framework to various categories of applications.

In privacy-preserving data publishing with publicly-known algorithms, we first theoretically study a generic strategy independent of data utility measures and syntactic privacy properties. We then propose an efficient approach to preserving diversity.

In privacy-preserving traffic padding in Web applications, we first propose a formal PPTP model to quantify the privacies and costs based on the key observation about the similarity between data publishing and traffic padding. We then introduce randomness into the previous solutions to provide background knowledge-resistant privacy guarantee.

In privacy-preserving smart metering, we propose a light-weight approach to simultaneously preserving privacy on both billing and consumption aggregation based on the key observation about the privacy issue beyond the fine-grained
readings.