On May 13th - 15th, 2020, we held an online workshop to share detailed plans for OpenDP and obtain community feedback on them. We covered topics such as the programming framework, governance, system integrations, use cases, statistical functionality, and collaborations. The content of the following documents were discussed during this workshop:
OpenDP Programming Framework, May 2020
Overview of OpenDP and Goals for Meeting
Speakers: Gary King, Open DP Faculty Director and Albert J. Weatherhead III University Professor, Director, IQSS, Harvard University
Salil Vadhan, OpenDP Faculty Director and Vicky Joseph Professor of Computer Science & Applied Mathematics, Harvard University
Annie Wu, Program Director for the Privacy Tools Project and OpenDP
We will give an overview of the plans and timeline for OpenDP and the goals for the workshop.
Use Cases
Speakers: Merce Crosas, University Research Data Management Officer, HUIT, Chief Data Science and Technology Officer, IQSS, Harvard University
Ilya Mironov, Research Scientist, Facebook AI
The goal of OpenDP is to help open up otherwise siloed and sequestered sensitive data to support scientifically oriented research and exploration in the public interest. To this end, we will discuss several concrete use cases for OpenDP, which include archival data repositories, government agencies, and companies sharing data with researchers.
Differential Privacy for COVID-19 Data
Speakers: Miguel Guevara, Product Manager, Google
Bryant Gipson, Engineering Manager, Google
We will give an overview of the three year journey and DP infrastructure that made it possible to launch Google's first-ever DP product (from the ground-up) in record time to address the Covid-19 crisis. We first present an overview of the reports, and continue by showing the different building blocks that are critical aspects of a robust DP infrastructure. We conclude with some insights and learnings we obtained.
Programming Framework for the DP Library
Speakers: Marco Gaboardi, Assistant Professor, Computer Science, Boston University
Michael Hay, Associate Professor, Computer Science, Colgate University
At the core of the OpenDP Commons will be a library of differentially private algorithms that will be contributed and vetted by the OpenDP community. We will propose a programming framework for this library, and discuss how it seeks to support extensibility, flexibility, programmability, modularity, usability, efficiency, and utility.
Statistical Functionality
Speaker: James Honaker, Chief Privacy Engineer and Research Associate, Privacy Tools Project and OpenDP, Harvard University
We will discuss what statistical methods and measures of statistical utility and validity we plan to prioritize and support in OpenDP. We will also cover what library and system components and features make for easier adoption by statisticians, data scientists and other applied analysts.
System Integrations
Speakers: James Honaker, Chief Privacy Engineer and Research Associate, Privacy Tools Project and OpenDP, Harvard University
Joshua Allen, Principal Data Scientist, Microsoft
For differential privacy software to be useful in applications, it must integrate with end-to-end systems, which may vary in their back-end storage and compute capabilities and in the needs and expertise of end users. We will give several examples of the kinds of systems we aim to support with OpenDP, including a system being developed in collaboration with Microsoft."
Governance and Licensing
Speakers: Chris Bavitz, WilmerHale Clinical Professor of Law, Harvard Law School and Managing Director, Cyberlaw Clinic at the Berkman Klein Center for Internet and Society
Annie Wu, Program Director for the Privacy Tools Project and OpenDP
We will describe our plans for the governance of OpenDP, including both community vetting of the software in the OpenDP Commons as well as a future transition to community governance of the project as a whole. Other issues to be discussed include the code of conduct and licensing.
Collaborations
Speakers: Merce Crosas, University Research Data Management Officer, HUIT, Chief Data Science and Technology Officer, IQSS, Harvard University
Sarah Bird, Principal Program Manager, Microsoft
We aim for OpenDP to have a wide array of partners in industry, government, and academia, including organizations that will contribute to or make use of the OpenDP software. We will discuss several models for how we can carry out such collaborations.
The Programming Framework for the DP Library
Discussion Leads:
Marco Gaboardi, Assistant Professor, Computer Science, Boston University
Michael Hay, Associate Professor, Computer Science, Colgate University
This breakout group will discuss the programming framework for the DP library as presented in the plenary session and detailed in a whitepaper to be distributed before the workshop. We seek to obtain feedback from DP researchers and software engineers who may contribute to the DP library, and those who have experience with algorithmic libraries in other software projects.