Abstract

Significant amount of individual information are being collected and analyzed today through a wide variety of applications across different industries. While pursuing better utility by discovering
knowledge from the data, an individual’s privacy may be compromised during an analysis: corporate networks monitor their online behavior, advertising companies collect and share their private
information, and cybercriminals cause financial damages through security breaches. To this end,
the data typically goes under certain anonymization techniques, e.g., CryptoPAn [Computer Networks’04], which replaces real IP addresses with prefix-preserving pseudonyms, or Differentially
Private (DP) [ICALP’06] techniques which modify the answer to a query by adding a zero-mean
noise distributed according to, e.g., a Laplace distribution. Unfortunately, most such techniques
either are vulnerable to adversaries with prior knowledge, e.g., some network flows in the data, or
require heavy data sanitization or perturbation, both of which may result in a significant loss of data
utility. Therefore, the fundamental trade-off between privacy and utility (i.e., analysis accuracy) has
attracted significant attention in various settings [ICALP’06, ACM CCS’14]. In line with this track
of research, in this dissertation we aim to build utility-maximized and privacy-preserving tools for
Internet communications. Such tools can be employed not only by dissidents and whistleblowers,
but also by ordinary Internet users on a daily basis. To this end, we combine the development of
practical systems with rigorous theoretical analysis, and incorporate techniques from various disciplines such as computer networking, cryptography, and statistical analysis. During the research,
we proposed three different frameworks in some well-known settings outlined in the following.

First, we propose The Multi-view Approach to preserve both privacy and utility in network trace
anonymization, Second, The R2DP Approach which is a novel technique on differentially private
mechanism design with maximized utility, and Third, The DPOD Approach that is a novel framework on privacy preserving Anomaly detection in the outsourcing setting.