New attacks against transformation-based privacy-preserving linear programming

Abstract

In this paper we demonstrate a number of attacks against proposed protocols for privacy-preserving linear programming, based on publishing and solving a transformed version of the problem instance. Our attacks exploit the geometric structure of the problem, which has mostly been overlooked in the previous analyses and is largely preserved by the proposed transformations. The attacks are efficient in practice and cast serious doubt to the viability of transformation-based approaches in general. © 2013 Springer-Verlag.