Abstract. With an increased focus on equity across the federal government, federal agencies and civil society have a need to join datasets with sensitive demographic data. Given legal, policy, and ethical constraints, the agencies that collect demographics cannot share them directly with such stakeholders in many cases. This presentation will show the application of Google's Private Join and Compute protocol to a public dataset of race and ethnicity information, enabling secure measurement of demographic disparities without explicit data sharing. We will describe the adaptations to open-source tools that were required to implement the protocol on government infrastructure. We will also illustrate potential use cases of the work that were ascertained through user interviews, including enabling third party audits of machine learning models. This presentation aims to be a primer on the practicalities of demonstrating a Privacy Enhancing Cryptography protocol on real infrastructure in a government setting.
Joint work with: Marina DeFrates, Samantha Weinstock
WPEC 2024: NIST Workshop on Privacy-Enhancing Cryptography 2024. Virtual, 2024-Sep-24–26.
NIST Workshop on Privacy-Enhancing Cryptography 2024
Starts: September 24, 2024Virtual
Security and Privacy: cryptography