Comparison of measuring information leakage for fully probabilistic systems

Quantifying implicit information leakage is important, especially for fully probabilistic systems (FPS). Although many quantitative methods have been proposed (including methods based on mutual information, on α-mutual information, on relative entropy and on pure probability, etc.), there has been little work analyzing their consistency and accuracy. In order to perform this analysis, these methods must be modeled with a uniform approach. In this paper, a light probabilistic process algebra (PPA-Lite) is presented, and some existing quantitative methods are uniformly characterized by using PPA-Lite. Further, their relationships are analyzed by proof and simulation, respectively. The results show that (1) most of methods concur in determining whether information is leaked; (2) the method based on (α-) mutual information is the most accurate if the distribution of the sent information is known. If not, the method based on relative entropy is the most accurate.