Differential Privacy of Big Data: An Overview

摘要

Differential privacy has seen dramatic development in recent decades as data mining of the statistical private datasets in a distributed big data environment has become an effective paradigm that, it is argued, guarantees the mathematically rigorous privacy of the participants by ensuring the equivalence of the analyzing results with the removal or addition of a single database item. However, challenges relating to the trade-off between privacy and utility still apply with the application of differential privacy. In this survey, we review and re-examine those new improvements of the differential privacy mainly in correlated scenarios, along with different methods of choosing the epsilon for achieving a better trade-off between the privacy and utility of the datasets in conventional settings, so as to build up deeper insights on specific technical aspects of this paradigm and its future trends of development.