Theory Seminar - Differential Privacy in Theory and Practice, Abie Flaxman

Explore differential privacy in theory and practice through this seminar by Abie Flaxman from the University of Washington. Delve into the application of differential privacy in the 2020 US Census, examining its impact on data accuracy and confidentiality. Learn about the constitutional requirements for the Decennial Census, the legal obligations for data protection, and the potential trade-offs between privacy and utility. Discover recent research findings on the implementation of differential privacy in census data, including a case study using 1940 census information. Gain insights into the challenges and opportunities presented by this new approach to data protection, and consider areas where theoretical innovations could lead to practical improvements. Understand the unique features of census data, including mandatory participation and legal confidentiality requirements. Explore the concept of differential privacy through concrete examples and discuss its potential to democratize decision-making regarding privacy and accuracy trade-offs in data collection and dissemination.

Intro
Acknowledgement and Disclosures
Outline
2010 Census: High-level Database Schema
Unique feature of this data 1: Participation is mandatory
Unique feature of this data 2: Title 13.9.2 of US Code requires confidentiality
What is Differential Privacy? (continued)
What is a database?
Concrete example: population counts by county
Convex Optimization to Make Noise Plausible
For first time this decade, a dip in King County's white population, census data shows
Why might e-DP in 2020 Census be a good thing? There is an opportunity to democratize tradeoff decision between privacy and accuracy