By Yan Huang
People have thought about innovative applications such as mining credit card transactions to identify fraud transaction, or querying databases of personal medical records for precision medical treatments. Obviously, combining data from several collaborative organizations generally improves the accuracy of the models being derived. On the other hand, these applications involve intensive processing of sensitive data that every organization wants to hold tight to themselves.
Secure computation has long been speculated to be a key technology for safely utilizing sensitive data owned by two or more distrustful parties. In this regard, exciting breakthroughs have taken place in the last decade, but most practical applications focused on the weaker honest-but-curious threat model (even though the protocols are expected to be executed in presence of fully malicious adversaries). In a recent Conference on Computer and Communications Security (CCS) paper by Zhu, Huang, and Cassel, the challenge of executing extreme-scale long-term secure computations as a on-demand service was studied.
A major finding of their work is a pool technique, which brings forward many indispensable advantages to make such service practical. The advantages include strong lifetime security guarantee of the established servers, nearly zero offline delay, memory-efficient scaling, support of reactive computations, and --- most importantly for practitioners who will build their future applications --- simple black-box programming interfaces. We instantiated the idea of pooling with JIMU, a state-of-the-art actively-secure computation schemes that was accepted to ASIACRYPT’17, and incorporated it into a software framework (download link). This work is supported by NSF award #1464113 and NIH award 1U01EB023685-01.