Abstract:
|
Formal privacy methodology such as differential privacy is changing the established framework of how we publish and share sensitive data and how we perform statistical inference under privacy constraints. Differential privacy proposed nearly fifteen years ago has seen an explosion in theoretical and practical developments across many data domains (e.g., health, financial, genomic, survey data), and the U.S. Census is planning differentially private releases of 2020 census data. But numerous technical and practical subtleties exist that limit differential privacy usability in statistical applications. At this roundtable we will continue discussions from the related invited session, "Private Data for the Public Good: Formal Privacy in Survey Organizations”, where invited speakers focused on formal privacy research agenda for complex surveys (John Abowd, US Census), impact on estimation of the cost of child care (Quentin Brummet, NORC), impact of formal privacy on social science research flow (Frauke Kreuter, University of Maryland), and ways to achieve optimal inference with privacy constrains (Aleksandra Slavkovic, Penn State). A subset of these speakers will join this roundtable.
|