chapter two

2 Differential privacy for machine learning (Part-1)

 

This chapter covers

  • Concept and the definition of differential privacy
  • Formulating widely adopted differential privacy mechanisms in use today that have served as the most important building blocks in various privacy-preserving algorithms and applications
  • Implementing various properties of differential privacy under different scenarios

We have investigated various privacy-related threats and vulnerabilities in machine learning and numerous concepts of privacy-enhancing technologies in the previous chapter. From now on, we will learn the details about several essential and popular privacy-enhancing technologies. The first one we will introduce here is differential privacy. Differential privacy is one of the most popular, yet influential privacy protection schemes used in many different applications today. It introduces a fascinating concept of making a dataset robust enough for the changes of any single sample in the dataset by computing the data statistics. For instance, we can consider machine learning models as complex statistics which describe the distribution of their training data. This chapter will walk you through our 1st part of differential privacy for machine learning.

We are now in the second chapter of this book. In this chapter, we will be looking at what differential privacy is and how it has been widely adopted in numerous practical applications, along with various essential properties of differential privacy.

2.1 What is Differential Privacy (DP)?

2.1.1 The Concept of Differential Privacy

2.1.2 How Differential Privacy Works?

2.2 Mechanisms of Differential Privacy

2.2.1 Binary Mechanism (Randomized Response)

2.2.2 Laplace Mechanism

2.2.3 Exponential Mechanism

2.3 Properties of Differential Privacy

2.3.1 Post-Processing Property of Differential Privacy

2.3.2 Group Privacy Property of Differential Privacy

2.3.3 Composition Properties of Differential Privacy

2.4 Summary