Secure Computation on Datasets

There are myriad domains where communities could benefit from aggregate statistical analyses that link across databases without requiring data owners to share their underlying data. In many cases, however, an obligation to protect the privacy of the underlying data often prevents organizations from performing (joint) statistical analyses that would benefit the community as a whole. Some common examples include: linking Electronic Medical Records (EMR) for longitudinal studies; student and teacher performance records; monitoring international financial transactions; environmental and hazard records; transportation and automotive studies; studies on logistics for managing vendor and supply chains; personal genetics and ancestry studies; cell phone call and location records; energy efficient HVAC and building management systems; digital statistics and cyber forensics.

In principle, general-purpose secure multiparty computation protocols exist that allow a group of mutually untrusting data owners to compute any function of interest across their joint data. This flexibility of functionality comes at a price, however, and existing MPC compilers target developers with a high-level of cryptographic expertise.  Our work focuses on building a simple, easy-to-deploy interface that allows non-crypto experts to securely compute a limited but optimized set of statistics on jointly held data.