Apparaît dans la collection : Nexus Trimester - 2016 - Central Workshop

Given two random variables X and Y, a new measure L(X;Y), called emph{G-leakage}, is proposed to quantify the amount of information that Y ‘‘leaks’’ about X. The measure is defined operationally as the multiplicative increase, upon observing Y, of the probability of correctly guessing a randomized function of X, maximized over all such randomized functions. G-leakage is inspired by both strong data processing inequalities in information theory and differential privacy in theoretical computer science, and it turns out to equal the Sibson mutual information of order infinity, endowing the latter with an operational significance. Moreover, it is shown that the definition is robust in several respects: it is unchanged even if it is modified to allow for several guesses or if the guess only needs to be within a certain distance of the true function value.