|
Activity Number:
|
239
- Synthetic Data and Differential Privacy: Data, Privacy and the Public Good
|
|
Type:
|
Invited
|
|
Date/Time:
|
Tuesday, August 4, 2020 : 1:00 PM to 2:50 PM
|
|
Sponsor:
|
Survey Research Methods Section
|
|
Abstract #309532
|
|
|
Title:
|
On the Formally Private Use of Public Historical Data to Control Upward Bias in Small Counts
|
|
Author(s):
|
Philip Leclerc*
|
|
Companies:
|
US Census Bureau
|
|
Keywords:
|
|
|
Abstract:
|
The 2020 Decennial Census will be the first nation-wide Census in the world to use formally private disclosure limitation methods to protect the privacy of respondents. The principal algorithm developed for this purpose is called the “Topdown Algorithm” (TDA). This algorithm proceeds recursively from the national level down to the Census block level. TDA generates fully saturated contingency tables (called “histograms” in the differential privacy literature) that are informationally equivalent to microdata. The corresponding microdata are used to support the production of tens of billions of tabulations about the demographic characteristics of the U.S. population. Many technical challenges were encountered in the development of TDA. In this talk, I discuss one important challenge: the lack of good theory for controlling upward bias. Upward bias in small counts results when using mathematical optimization to estimate nonnegative integer-valued histograms at massive scale. To combat this problem, we have developed techniques that use public historical data to estimate “small cell structure” without weakening differential privacy’s worst-case privacy guarantees. I will discuss bo
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2020 program
|
