William Sexton, Tumult Labs
The Economic Programs Directorate of the U.S. Census Bureau releases several data products detailing statistics about establishments that need to be protected using privacy methods like differential privacy (DP). However, due to the highly skewed nature of the data, standard DP algorithms permit little or no useful data to be released while ensuring the same level of privacy for all establishments. We present a new formal privacy framework, Per-Record Differential Privacy (PRDP) that provides a sliding scale privacy guarantees with small establishments receiving stronger privacy protections and large establishments receiving weaker privacy protections. We will discuss algorithms for an exemplar data product under PRDP and outline the advantages and limitations of this new approach to privacy for skewed data. This new privacy methodology was recently deployed as part of a demonstration data product for the Census Bureau's County Business Patterns data product.
William Sexton, Tumult Labs
William Sexton is a scientist at Tumults Labs, where he has contributed to the development of privacy algorithms for several projects with the U.S. Census Bureau including County Business Patterns and the Detailed and Supplemental - Demographic and Housing Characteristics data products. Prior to joining Tumult, he was a researcher and developer at the US Census Bureau where he was part of the team that designed and implemented large-scale privacy technologies for the 2020 decennial census. He received his Ph.D. in Economics from Cornell University. He received a Bachelor's degree and a Master's degree in Mathematics from Brigham Young University.
author = {William Sexton},
title = {{Per-Record} Privacy and Its Application to {Heavy-Tailed} Economic Data},
year = {2023},
address = {Santa Clara, CA},
publisher = {USENIX Association},
month = sep
}