We define information leakage in terms of a “difference” between the a priori distribution over some remote behavior and the a posteriori distribution of the remote behavior conditioned on a local observation from a protocol run. Either a maximum or an average may be used. We identify a set of notions of “difference;” we show that they reduce our general leakage notion to various definitions in the literature. We also prove general composability theorems analogous to the data-processing inequality for mutual information, or cascading channels for channel capacities.
Ando, M., & Guttman, J. D. (2016). Composable bounds on information flow from distribution differences. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9481, pp. 13–29). Springer Verlag. https://doi.org/10.1007/978-3-319-29883-2_2