Concrete attacks against existing privacy-preserving linear programming methods
Kadri Tõldsepp

While in a recent news item, we told about weaknesses we had found in existing transformation-based privacy-preserving linear programming methods, we can now report that based on these weaknesses, we have constructed concrete attacks against several methods.
 While the analyses by the authors of various methods have mostly concentrated on the algebraic properties of the transformations, we take a closer look at the geometric properties of the original and the transformed feasible region. We show how the bounding hyperplanes of these regions can be matched against each other, leading to the discovery of the private parameters of the transformation.