Hands-On Differential Privacy

Working with sensitive data means facing a difficult question: how can useful analysis be shared without exposing what any one person contributed? Hands-On Differential Privacy tackles that problem from both sides—explaining the underlying privacy guarantees and showing how to put them to work with practical tools, including OpenDP.

Ethan Cowan, Michael Shoemate, and Mayana Pereira lead readers from foundational ideas to statistical and machine-learning applications, with examples that make the trade-offs between privacy and analytical utility part of the conversation.

Start with the privacy guarantees

The book introduces privacy units, sensitivity, randomized response, noise mechanisms, composition, and the formal models used to describe differential privacy. Rather than treating privacy as a vague promise, it examines what a guarantee means, how it can be interpreted, and what choices affect its strength.

Connect the mathematics to working methods

Practical chapters explore differentially private queries, stable transformations, data truncation, privacy amplification, and analysis involving unbounded contributions. Examples and exercises help readers follow how the concepts shape real analytical workflows, including the use of the OpenDP library.

From statistics to machine learning

The book extends its discussion to private statistical modeling, differentially private machine learning, and synthetic data. It also considers privacy attacks and reidentification risks, bringing attention to why careful implementation and interpretation matter—not just the selection of an algorithm.

For practitioners who need more than a definition

Data scientists, researchers, programmers, and engineers working with sensitive datasets will find a technically oriented introduction that joins theory with applied examples. Readers new to differential privacy can build a conceptual foundation; those already working in the area can use the material to examine mechanisms, guarantees, and deployment considerations in greater depth.

For anyone responsible for producing useful analysis while respecting individual privacy, this book offers a structured way to understand both the promise of differential privacy and the decisions involved in applying it.