The companion SURVEY ARTICLE: How to DP-fy Your Data: A Practical Guide to Generating Synthetic Data With Differential Privacy
is available here: https://arxiv.org/abs/2512.03238. The article offers an extended treatment of the topics covered in this tutorial.
|
📹 RECORDING
6 sections, 2h15m |
📄 SLIDES
PDF, 9.7MB download |
💻 DEMO
Generate DP synthetic data in Colab |
ABSTRACT
This tutorial focuses on differentially private (DP) synthetic data generation. Creating DP synthetic data allows for data sharing without compromising individuals' privacy, opening up possibilities for collaborative model development. This tutorial provides a comprehensive guide for generating DP synthetic data across text, image, and tabular modalities.
This tutorial covers: an introduction to synthetic data and differential privacy; specific generation methods for various data types (text, image, and tabular); as well as practical aspects of real-world deployments such as user-level guarantees, empirical privacy testing, and data lineage.
CONTACT
For any questions or comments about anything here, please email us: nponomareva@google.com, sergeiv@google.com, kairouz@google.com, alexbie@google.com.