A Least Squares approach to user profiling in pool mix-based anonymous communication systems

Deployed high-latency anonymous communication systems conceal communication patterns using pool mixes as building blocks. These mixes are known to be vulnerable to Disclosure Attacks that uncover persistent relationships between users. In this paper we study the performance of the Least Squares Disclosure Attack (LSDA), an approach to disclosure rooted in Maximum Likelihood parameter estimation that recovers user profiles with greater accuracy than previous work. We derive analytical expressions that characterize the profiling error of the LSDA with respect to the system parameters for a threshold binomial pool mix and validate them empirically. Moreover, we show that our approach is easily adaptable to attack diverse pool mixing strategies.