Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Synthetic Data -- what, why and how?

James Jordon, Lukasz Szpruch, Florimond Houssiau, Mirko Bottarelli, Giovanni Cherubin, Carsten Maple, Samuel N. Cohen, Adrian Weller

TL;DR

The paper surveys synthetic data technologies with a focus on privacy, outlining formal notions like differential privacy, threat models, and attacks such as membership and reconstruction. It emphasizes that synthetic data is not automatically private and is not a universal substitute for real data, outlining utility, fidelity, and privacy trade-offs. It discusses auditing practices, de-biasing approaches, and data augmentation, and highlights industry perspectives on adoption and governance. Overall, it argues for careful, use-case-driven deployment of synthetic data, with rigorous evaluation and context-aware privacy guarantees to realize its benefits while mitigating risks.

Abstract

This explainer document aims to provide an overview of the current state of the rapidly expanding work on synthetic data technologies, with a particular focus on privacy. The article is intended for a non-technical audience, though some formal definitions have been given to provide clarity to specialists. This article is intended to enable the reader to quickly become familiar with the notion of synthetic data, as well as understand some of the subtle intricacies that come with it. We do believe that synthetic data is a very useful tool, and our hope is that this report highlights that, while drawing attention to nuances that can easily be overlooked in its deployment.

Synthetic Data -- what, why and how?

TL;DR

The paper surveys synthetic data technologies with a focus on privacy, outlining formal notions like differential privacy, threat models, and attacks such as membership and reconstruction. It emphasizes that synthetic data is not automatically private and is not a universal substitute for real data, outlining utility, fidelity, and privacy trade-offs. It discusses auditing practices, de-biasing approaches, and data augmentation, and highlights industry perspectives on adoption and governance. Overall, it argues for careful, use-case-driven deployment of synthetic data, with rigorous evaluation and context-aware privacy guarantees to realize its benefits while mitigating risks.

Abstract

This explainer document aims to provide an overview of the current state of the rapidly expanding work on synthetic data technologies, with a particular focus on privacy. The article is intended for a non-technical audience, though some formal definitions have been given to provide clarity to specialists. This article is intended to enable the reader to quickly become familiar with the notion of synthetic data, as well as understand some of the subtle intricacies that come with it. We do believe that synthetic data is a very useful tool, and our hope is that this report highlights that, while drawing attention to nuances that can easily be overlooked in its deployment.
Paper Structure (76 sections, 4 equations, 1 figure)

This paper contains 76 sections, 4 equations, 1 figure.

Table of Contents

  1. Synthetic data is a technology with significant promise.
  2. Synthetic data can accelerate development.
  3. Synthetic data is not automatically private.
  4. Synthetic data is not a replacement for real data.
  5. Outliers are hard to capture privately.
  6. Empirically evaluating the privacy of a single dataset can be problematic.
  7. Black box models can be particularly opaque when it comes to generating synthetic data.
  8. Synthetic data goes beyond privacy.
  9. Introduction
  10. Report Structure
  11. What is Synthetic Data?
  12. Can synthetic data replace real data?
  13. How should we approach doing things with synthetic data?
  14. Data Linking
  15. Combining Synthetic Data with Other Technologies
  16. ...and 61 more sections

Figures (1)

  • Figure 1: Generative modelling taxonomy based on the maximum likelihood from goodfellow_nips_2017

Theorems & Definitions (4)

  • Definition 1
  • Definition 2: Differential Privacy dpbook
  • Definition 3: Fairness Through Unawareness decaf
  • Definition 4: Demographic Parity decaf