chapter nine

9 Compressive privacy for machine learning

 

This chapter covers

  • Understanding compressive privacy
  • Introducing compressive privacy for machine learning applications
  • Implementing compressive privacy from theory to practice
  • A compressive privacy solution for privacy-preserving machine learning

In the previous chapters, we looked into the concept of differential privacy, local differential privacy, privacy-preserving synthetic data generation, privacy-preserving data mining, and their applications in designing privacy-preserving machine learning solutions. In a quick recap, in differential privacy, a trusted data curator collects data from individuals and produces differentially private results by adding precisely computed noise to the aggregation of individuals’ data. In local differential privacy, individuals send their data to the data aggregator after privatizing data by perturbation, which eliminates the need for a trusted data curator to collect the private data from individuals. Privacy-preserving synthetic data generation provides a promising solution for private data sharing, which generates synthetic yet representative data that can be used to share among multiple parties safely and securely.

9.1 Introduction to Compressive Privacy

9.2 The Mechanisms of Compressive Privacy

9.2.1 Principal Component Analysis (PCA)

9.2.2 Other Dimensionality Reduction (DR) Methods

9.3 Implementing Compressive Privacy for Machine Learning Applications

9.3.1 The Accuracy of the Utility Task

9.3.2 The Effect of ρ' in DCA for Privacy and Utility

EXERCISE 01: TRY IT YOURSELF

EXERCISE 02: TRY IT YOURSELF

EXERCISE 03: TRY IT YOURSELF

9.4 Case Study: Privacy-Preserving PCA/DCA on Horizontally Partitioned Data

HOW TO ACHIEVE PRIVACY-PRESERVATION ON HORIZONTALLY PARTITIONED DATA

9.4.1 Recap on Different Dimensionality Reduction Approaches

9.4.2 Use of Additive Homomorphic Encryption

9.4.3 Overview of the Proposed Approach

9.4.4 How Privacy-Preserving Computation Works

PRIVACY-PRESERVING PCA

PRIVACY-PRESERVING DCA

9.4.5 Evaluating the Efficiency and Accuracy of the Privacy Preserving PCA/DCA

ANALYZING THE EFFICIENCY

ANALYZING THE ACCURACY OF THE ML TASK

9.5 Summary

REFERENCES