Date Published: December 11, 2023
Comments Due:
Email Comments to:
Author(s)
Joseph Near (University of Vermont), David Darais (Galois)
Editor(s)
Naomi Lefkovitz (NIST), Gary Howarth (NIST)
Announcement
This publication is about differential privacy, a privacy-enhancing technology that quantifies privacy risk to individuals when their information appears in a dataset. In response to President Biden’s Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, SP 800-226 is intended to help agencies and practitioners of all backgrounds—policy makers, business owners, product managers, IT technicians, software engineers, data scientists, researchers, and academics—better understand how to evaluate promises made (and not made) when deploying differential privacy, including for privacy-preserving machine learning. Additionally, there is a supplemental package of Python Jupyter notebooks that illustrate how to achieve differential privacy and other concepts described in the publication.
Submit comments by 11:59 p.m. EST on Thursday, January 25, 2024 to privacyeng@nist.gov. We encourage you to use this comment template.
Note to Reviewers
The authors welcome feedback on all aspects of this publication, particularly on the following questions:
- Does this publication have a clear and appropriate scope?
- Is this publication understandable for the intended audience?
- Does publication provide a conceptual framework for understanding the uses and pitfalls of differential privacy? Is there any guidance that is not well-founded?
- Is the differential privacy pyramid a helpful conceptual device?
- Are the privacy hazards described accurately? Should additional hazards be added?
- For topics where the research is inconclusive, were any key points missed from the literature?
NOTE: A call for patent claims is included on page ii of this draft. For additional information, see the Information Technology Laboratory (ITL) Patent Policy – Inclusion of Patents in ITL Publications.
This publication describes differential privacy — a mathematical framework that quantifies privacy risk to individuals as a consequence of data collection and subsequent data release. It serves to fulfill one of the assignments to the National Institute of Standards and Technology (NIST) by the Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence issued on October 30, 2023. The primary goal of this publication is to help practitioners of all backgrounds better understand how to think about differentially private software solutions. Multiple factors for consideration are identified in a differential privacy pyramid along with several privacy hazards, which are common pitfalls that arise as the mathematical framework of differential privacy is realized in practice.
This publication describes differential privacy — a mathematical framework that quantifies privacy risk to individuals as a consequence of data collection and subsequent data release. It serves to fulfill one of the assignments to the National Institute of Standards and Technology (NIST) by the...
See full abstract
This publication describes differential privacy — a mathematical framework that quantifies privacy risk to individuals as a consequence of data collection and subsequent data release. It serves to fulfill one of the assignments to the National Institute of Standards and Technology (NIST) by the Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence issued on October 30, 2023. The primary goal of this publication is to help practitioners of all backgrounds better understand how to think about differentially private software solutions. Multiple factors for consideration are identified in a differential privacy pyramid along with several privacy hazards, which are common pitfalls that arise as the mathematical framework of differential privacy is realized in practice.
Hide full abstract
Keywords
anonymization; data analytics; data privacy; de-identification; differential privacy; privacy; privacy-enhancing technologies
Control Families
None selected
) or https:// don't prove our archive is authentic, only that you securely accessed it. Note that we are working to fix that :)
