Computing Statistics from Private Data
This project will lay a foundation for the development of cryptographic tools from the field of secure Multiparty Computation (MPC) to perform joint statistical analyses across multiple private data sets, without the need to share the underlying data.
We plan two workshops that will bring together social scientists and statisticians experienced in working with administrative data with computer scientists and cryptographers who are experts in MPC. We will focus on three types of administrative data and identify requirements for an MPC system tailored to generate standard statistical analyses in a secure and efficient way. We will also discuss acceptable threat models and leakage levels. Previous work has found that leaking partial information (e.g. the size of data sets in an intermediate computation) allowed for significant flexibility in protocol design and ultimately led to large efficiency gains. Results of these workshops will be: (1) A list of the statistical procedures the MPC system will perform and the order in which they will be implemented during software development. (2) A security model and guidelines for acceptable leakage levels. (3) A short-list of the MPC protocols that are most likely to be efficient with our target data types. These outcomes will be incorporated in an NSF proposal to build and test an MPC system.
Funding: Alfred P. Sloan Foundation (G-2014-13803)
Funding Period: 1/1/2015 to 12/31/2016