I will describe our recent experience, building an FHE-based system for computing the coefficients of an approximate logistic-regression model. The aim of this project was to examine the feasibility of a solution that operates "deep within the bootstrapping parameter regime", solving a complicated system that cannot be addressed just by using a somewhat homomorphic scheme. Our solution can handle thousands of records and hundreds of fields, and it takes a few hours to run. In this presentation I will talk about the challenges of designing and implementing this solution, and about some of the optimizations that went into making it feasible.

Time  permitting, I will also talk about a number of other recent optimizations that we developed for homomorphic packed linear transformations (even though we ended up not using any of them in this project).

based on joint work with Jack Lik Hon Crawford, Craig Gentry, Daniel Platt, and Victor Shoup