Abstract

Chapter three key characteristics of programming frameworks for differential privacy: privacy calculus, which is how frameworks provide quantitative methods for bounding privacy loss without the need to compute probability distributions explicitly (Section 12.3), composition and interactivity, discussed in Section 12.4, which is how frameworks handle the cumulative privacy loss over multiple analyses, and expressivity, discussed in Section 12.5, refers to reviews the richness and variety of analyses that frameworks can express. In addition to these core characteristics, we discuss tools and techniques for the verification and testing of differential privacy implementations (Section 12.6). .