Cover Image

Differential Privacy in Artificial Intelligence From, Theory to Practice.

The ebook edition of this title is Open Access and freely available to read online. This book delves into the theoretical underpinnings of differential privacy, its use in machine learning systems, practical implementation details, and its broader social and legal ramifications.

Bibliographic Details
Main Author: Fioretto, Ferdinando
Corporate Author: Emerald Publishing Group
Other Authors: Van Hentenryck, Pascal
Format: eBook
Language:English
Published: Norwell, MA : Now Publishers, 2025.
Subjects:
Data privacy.
Artificial intelligence.
Confidentialité des données.
Intelligence artificielle.
artificial intelligence.
Computers > Internet > Online Safety & Privacy.
Online safety and behaviour.
Privacy and data protection.
Electronic books.
Online Access:Connect to the full text of this electronic book
Table of Contents:
  • Cover
  • DIFFERENTIAL PRIVACY IN ARTIFICIAL INTELLIGENCE: FROM, THEORY TO PRACTICE
  • Copyright
  • Table of Contents
  • I: Foundation
  • 1: Overview and Fundamental Techniques
  • 1.1 Introduction
  • 1.2 A Historical Perspective on Privacy
  • 1.2.1 Data Anonymization
  • Why Did Anonymization Fail?
  • 1.2.2 K-Anonymity
  • Where Does k-anonymization Fail? Reason #1: Lack of Group Privacy
  • Where Does k-anonymization Fail? Reason #2: Lack of Composition
  • 1.2.3 Any Perfectly Accurate and Deterministic Privacy Notion Must Fail
  • 1.2.4 A Side Note: Other Types of Privacy Breaches
  • 1.3 What Protections Does Differential Privacy Provide?
  • 1.3.1 What Does Differential Privacy Promise?
  • First Attempt: No Information Leakage
  • Second Attempt: Almost No Information Leakage
  • Refining the Definition of Privacy
  • 1.3.2 Where to Guarantee Differential Privacy? Local vs Central Models
  • Distinguishing Data Privacy From Data Security
  • 1.4 Differential Privacy: Formal Definition, Techniques, and Properties
  • Randomized Response
  • 1.4.1 Differential Privacy, Formally
  • Datasets and Queries
  • Global Sensitivity
  • Differential Privacy
  • 1.4.2 Formal Properties of Differential Privacy
  • Composition
  • Group Privacy
  • Post-processing
  • Quantifiable Privacy-accuracy Trade-offs
  • 1.4.3 The Laplace Mechanism
  • Accuracy Guarantee of the Laplace Mechanism
  • 1.4.4 Answering Private Queries in Practice
  • Example 1: Computing the Average Age
  • Example 2: Releasing a Histogram
  • 1.5 Approximate Differential Privacy
  • 1.5.1 The Gaussian Mechanism
  • Discussion of Accuracy
  • 1.6 Beyond Statistical Queries: Differentially Private Selection
  • 1.6.1 The Exponential Mechanism
  • Accuracy Guarantee
  • 1.7 Randomized Response, Revisited
  • Revisiting Randomized Response
  • Privacy Guarantees
  • Accuracy of Randomized Response
  • 1.8 Concluding Remarks
  • Acknowledgements
  • References
  • 2: Local Differential Privacy for Privacy-preserving Machine Learning
  • 2.1 Introduction
  • 2.1.1 A First LDP Protocol
  • 2.1.2 A Brief History of LDP
  • Overview of the Chapter
  • 2.2 Randomized Response
  • 2.3 Frequency Oracles
  • 2.3.1 Direct Encoding
  • 2.3.2 Unary Encoding
  • 2.3.3 Hash Encoding
  • 2.3.4 Hadamard Encoding
  • 2.3.5 Domain Size Reduction
  • 2.4 Heavy Hitters, Marginals, and Range Queries
  • 2.4.1 Heavy Hitters
  • 2.4.2 Marginals
  • 2.4.3 Range Queries and Quantiles
  • 2.5 Local Differential Privacy in Applications
  • 2.5.1 Text and Language Modeling
  • 2.5.2 Spatial Data
  • 2.5.3 Graphs and Social Network Data
  • 2.5.4 Classification and Regression
  • Minimizing a Loss Function
  • Vector Release
  • Putting it All Together
  • 2.5.5 Recommender Systems
  • 2.5.6 Common Themes for LDP in Machine Learning Applications
  • Finding a Good Class of Models
  • Collect Data That can be Combined Linearly
  • Reduce to Well-understood Problems
  • Noise Reduction Techniques
  • Data Representation