Artificial Intelligence and Differential Privacy : Review of Protection Estimate Models

Abstract

Differential Privacy (DP) can provide strong guarantees that personal information is not disclosed in data sets. This is ensured from mathematical, theoretical, and relational proof of privacy, which makes it important to understand the actual behavior of the DP-based protection models. For this purpose, we will review what kind of frameworks or models are available to estimate how well an implemented differential privacy model works. Special attention is paid to how to assess that a certain level of privacy has been reached, what configurations were used, and how to estimate the privacy loss. Our goal is to locate a common framework that could help one decide, based on privacy requirements, which model and configuration should be used and how its protection can be ensured.