Differentially private applications: Where to start?

Abstract

A lot of differentially private applications have been proposed nowadays. The various steps that can be followed when solving a privacy preservation problem for a particular application are shown in the first figure of this chapter. The dark boxes in the flowchart show the steps, and the orange boxes illustrate the possible choices. First, it is necessary to identify the scenarios: data publishing or data analysis. Data publishing aims to release answers to queries or entire datasets to public users; whereas, data analysis normally releases a private version of a model. Because private learning frameworks solve privacy preservation problems using optimization, an optimization objective normally has to be determined. The second step is identifying challenges in the application. Although differential privacy is considered to be a promising solution for privacy preservation issues, implementation in some applications still presents a number of challenges. These challenges, and their possible solutions, are introduced in the next subsection.