PINQuin, a framework for differentially private analysis

Abstract

Privacy is a humans right to seclude themselves, or information about themselves, from surveillance and view of others. Different cultures define and respect privacy differently but they usually share some basic concepts. Usually a special information that distincts one person or a group of population from others is considered personally sensitive. One aspect of privacy is related to anonymity which tries to remove or hide this personally sensitive information. Differential privacy is a robust standard that aims to protect individual's privacy when disclosing results from statistical analysis. In this master thesis we present a new model for differentially private data mining. As a result of this thesis we introduce a new method for privacy budgeting, namely record based differential privacy budgeting and PINQuin, our framework based on PINQ, that uses this new method. We also present experimental results comparing PINQuin with PINQ.