Vida: How to use bayesian inference to de-anonymize persistent communications

Abstract

We present the Vida family of abstractions of anonymous communication systems, model them probabilistically and apply Bayesian inference to extract patterns of communications and user profiles. The first is a very generic Vida Black-box model that can be used to analyse information about all users in a system simultaneously, while the second is a simpler Vida Red-Blue model, that is very efficient when used to gain information about particular target senders and receivers. We evaluate the Red-Blue model to find that it is competitive with other established long-term traffic analysis attacks, while additionally providing reliable error estimates, and being more flexible and expressive. © 2009 Springer Berlin Heidelberg.