book

Privacy-Preserving Machine Learning

Name: Privacy-Preserving Machine Learning
ISBN: 9781617298042

by Dumindu Samaraweera, Morris Chang, Di Zhuang

May 2023

Intermediate to advanced

336 pages

10h 3m

English

Manning Publications

Audiobook available

Read now

Unlock full access

inside front cover
Privacy-Preserving Machine Learning
Copyright
contents
front matter
prefaceacknowledgmentsabout this bookWho should read this bookHow this book is organized: A road mapAbout the codeliveBook discussion forumabout the authorsabout the cover illustration
Part 1 Basics of privacy-preserving machine learning with differential privacy
1 Privacy considerations in machine learning
1.1 Privacy complications in the AI era1.2 The threat of learning beyond the intended purpose1.2.1 Use of private data on the fly1.2.2 How data is processed inside ML algorithms1.2.3 Why privacy protection in ML is important1.2.4 Regulatory requirements and the utility vs. privacy tradeoff1.3 Threats and attacks for ML systems1.3.1 The problem of private data in the clear1.3.2 Reconstruction attacks1.3.3 Model inversion attacks1.3.4 Membership inference attacks1.3.5 De-anonymization or re-identification attacks1.3.6 Challenges of privacy protection in big data analytics1.4 Securing privacy while learning from data: Privacy-preserving machine learning1.4.1 Use of differential privacy1.4.2 Local differential privacy1.4.3 Privacy-preserving synthetic data generation1.4.4 Privacy-preserving data mining techniques1.4.5 Compressive privacy1.5 How is this book structured?Summary
2 Differential privacy for machine learning
2.1 What is differential privacy?2.1.1 The concept of differential privacy2.1.2 How differential privacy works2.2 Mechanisms of differential privacy2.2.1 Binary mechanism (randomized response)2.2.2 Laplace mechanism2.2.3 Exponential mechanism2.3 Properties of differential privacy2.3.1 Postprocessing property of differential privacy2.3.2 Group privacy property of differential privacy2.3.3 Composition properties of differential privacySummary
3 Advanced concepts of differential privacy for machine learning
3.1 Applying differential privacy in machine learning3.1.1 Input perturbation3.1.2 Algorithm perturbation3.1.3 Output perturbation3.1.4 Objective perturbation3.2 Differentially private supervised learning algorithms3.2.1 Differentially private naive Bayes classification3.2.2 Differentially private logistic regression3.2.3 Differentially private linear regression3.3 Differentially private unsupervised learning algorithms3.3.1 Differentially private k-means clustering3.4 Case study: Differentially private principal component analysis3.4.1 The privacy of PCA over horizontally partitioned data3.4.2 Designing differentially private PCA over horizontally partitioned data3.4.3 Experimentally evaluating the performance of the protocolSummary
Part 2 Local differential privacy and synthetic data generation

4 Local differential privacy for machine learning
4.1 What is local differential privacy?4.1.1 The concept of local differential privacy4.1.2 Randomized response for local differential privacy4.2 The mechanisms of local differential privacy4.2.1 Direct encoding4.2.2 Histogram encoding4.2.3 Unary encodingSummary
5 Advanced LDP mechanisms for machine learning
5.1 A quick recap of local differential privacy5.2 Advanced LDP mechanisms5.2.1 The Laplace mechanism for LDP5.2.2 Duchi’s mechanism for LDP5.2.3 The Piecewise mechanism for LDP5.3 A case study implementing LDP naive Bayes classification5.3.1 Using naive Bayes with ML classification5.3.2 Using LDP naive Bayes with discrete features5.3.3 Using LDP naive Bayes with continuous features5.3.4 Evaluating the performance of different LDP protocolsSummary
6 Privacy-preserving synthetic data generation
6.1 Overview of synthetic data generation6.1.1 What is synthetic data? Why is it important?6.1.2 Application aspects of using synthetic data for privacy preservation6.1.3 Generating synthetic data6.2 Assuring privacy via data anonymization6.2.1 Private information sharing vs. privacy concerns6.2.2 Using k-anonymity against re-identification attacks6.2.3 Anonymization beyond k-anonymity6.3 DP for privacy-preserving synthetic data generation6.3.1 DP synthetic histogram representation generation6.3.2 DP synthetic tabular data generation6.3.3 DP synthetic multi-marginal data generation6.4 Case study on private synthetic data release via feature-level micro-aggregation6.4.1 Using hierarchical clustering and micro-aggregation6.4.2 Generating synthetic data6.4.3 Evaluating the performance of the generated synthetic dataSummary
Part 3 Building privacy-assured machine learning applications
7 Privacy-preserving data mining techniques
7.1 The importance of privacy preservation in data mining and management7.2 Privacy protection in data processing and mining7.2.1 What is data mining and how is it used?7.2.2 Consequences of privacy regulatory requirements7.3 Protecting privacy by modifying the input7.3.1 Applications and limitations7.4 Protecting privacy when publishing data7.4.1 Implementing data sanitization operations in Python7.4.2 k-anonymity7.4.3 Implementing k-anonymity in PythonSummary
8 Privacy-preserving data management and operations
8.1 A quick recap of privacy protection in data processing and mining8.2 Privacy protection beyond k-anonymity8.2.1 l-diversity8.2.2 t-closeness8.2.3 Implementing privacy models with Python8.3 Protecting privacy by modifying the data mining output8.3.1 Association rule hiding8.3.2 Reducing the accuracy of data mining operations8.3.3 Inference control in statistical databases8.4 Privacy protection in data management systems8.4.1 Database security and privacy: Threats and vulnerabilities8.4.2 How likely is a modern database system to leak private information?8.4.3 Attacks on database systems8.4.4 Privacy-preserving techniques in statistical database systems8.4.5 What to consider when designing a customizable privacy-preserving database systemSummary
9 Compressive privacy for machine learning
9.1 Introducing compressive privacy9.2 The mechanisms of compressive privacy9.2.1 Principal component analysis (PCA)9.2.2 Other dimensionality reduction methods9.3 Using compressive privacy for ML applications9.3.1 Implementing compressive privacy9.3.2 The accuracy of the utility task9.3.3 The effect of ρ' in DCA for privacy and utility9.4 Case study: Privacy-preserving PCA and DCA on horizontally partitioned data9.4.1 Achieving privacy preservation on horizontally partitioned data9.4.2 Recapping dimensionality reduction approaches9.4.3 Using additive homomorphic encryption9.4.4 Overview of the proposed approach9.4.5 How privacy-preserving computation works9.4.6 Evaluating the efficiency and accuracy of the privacy-preserving PCA and DCASummary
10 Putting it all together: Designing a privacy-enhanced platform (DataHub)
10.1 The significance of a research data protection and sharing platform10.1.1 The motivation behind the DataHub platform10.1.2 DataHub’s important features10.2 Understanding the research collaboration workspace10.2.1 The architectural design10.2.2 Blending different trust models10.2.3 Configuring access control mechanisms10.3 Integrating privacy and security technologies into DataHub10.3.1 Data storage with a cloud-based secure NoSQL database10.3.2 Privacy-preserving data collection with local differential privacy10.3.3 Privacy-preserving machine learning10.3.4 Privacy-preserving query processing10.3.5 Using synthetic data generation in the DataHub platformSummary
Appendix A. More details about differential privacy
A.1 The formal definition of differential privacyA.2 Other differential privacy mechanismsA.2.1 Geometric mechanismA.2.2 Gaussian mechanismA.2.3 Staircase mechanismA.2.4 Vector mechanismA.2.5 Wishart mechanismA.3 Formal definitions of composition properties of DPA.3.1 The formal definition of sequential composition DPA.3.2 The formal definition of parallel composition DP
references
Appendix
index
inside back cover

Overview

Keep sensitive user data safe and secure without sacrificing the performance and accuracy of your machine learning models.

In Privacy Preserving Machine Learning, you will learn:

Privacy considerations in machine learning
Differential privacy techniques for machine learning
Privacy-preserving synthetic data generation
Privacy-enhancing technologies for data mining and database applications
Compressive privacy for machine learning

Privacy Preserving Machine Learning is a comprehensive guide to avoiding data breaches in your machine learning projects. You’ll get to grips with modern privacy-enhancing techniques such as differential privacy, compressive privacy, and synthetic data generation. Based on years of DARPA-funded cybersecurity research, ML engineers of all skill levels will benefit from incorporating these privacy-preserving practices into their model development. By the time you’re done reading, you’ll be able to create machine learning systems that preserve user privacy without sacrificing data quality and model performance.

About the Technology
Machine learning applications need massive amounts of data. It’s up to you to keep the sensitive information in those data sets private and secure. Privacy preservation happens at every point in the ML process, from data collection and ingestion to model development and deployment. This practical book teaches you the skills you’ll need to secure your data pipelines end to end.

About the Book
Privacy Preserving Machine Learning explores privacy preservation techniques through real-world use cases in facial recognition, cloud data storage, and more. You’ll learn about practical implementations you can deploy now, future privacy challenges, and how to adapt existing technologies to your needs. Your new skills build towards a complete security data platform project you’ll develop in the final chapter.

What's Inside

Differential and compressive privacy techniques
Privacy for frequency or mean estimation, naive Bayes classifier, and deep learning
Privacy-preserving synthetic data generation
Enhanced privacy for data mining and database applications

About the Reader
For machine learning engineers and developers. Examples in Python and Java.

About the Authors
J. Morris Chang is a professor at the University of South Florida. His research projects have been funded by DARPA and the DoD. Di Zhuang is a security engineer at Snap Inc. G. Dumindu Samaraweera is an assistant research professor at the University of South Florida. The technical editor for this book, Wilko Henecka, is a senior software engineer at Ambiata where he builds privacy-preserving software.

Quotes
A detailed treatment of differential privacy, synthetic data generation, and privacy-preserving machine-learning techniques with relevant Python examples. Highly recommended!
- Abe Taha, Google

A wonderful synthesis of theoretical and practical. This book fills a real need.
- Stephen Oates, Allianz

The definitive source for creating privacy-respecting machine learning systems. This area in data-rich environments is so important to understand!
- Mac Chambers, Roy Hobbs Diamond Enterprises

Covers all aspects for data privacy, with good practical examples.
- Vidhya Vinay, Streamingo Solutions

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

Publisher Resources

ISBN: 9781617298042Publisher Support Publisher Website

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills