CS Talk - Mahdi Zamani
Title: Reliable and Scalable Anonymity
Host: Bryan Ford
Abstract: Today, mass surveillance and censorship severely threaten democracy and freedom of speech. This makes anonymity a crucial tool for preserving privacy of individuals in cyberspace. Unfortunately, popular anonymity systems such as Tor cannot guarantee availability, reliability, or resistance to traffic-analysis attacks. In this talk, I address these limitations by designing randomized algorithms that can baffle the adversary with overwhelming probability.
As my first goal, I consider the problem of reliable bridge distribution in Tor. When access to the Tor network is blocked, Tor users have the option to use bridges -- volunteer relay nodes that are not listed in Tor's public directory. Bridge addresses are carefully distributed to the users, with the hope that they are not all learned by censors. A current challenge is that censors use a coalition of corrupt users to obtain bridge addresses and thus block a large number of bridges. We describe a randomized bridge distribution algorithm, where all (honest) users are guaranteed to be able to connect to Tor in the presence of an adversary corrupting an unknown number of users T. Our algorithm adaptively increases the number of bridges according to the behavior of the adversary and requires soft-O(T) bridges. Using our algorithm, the number of times a user fails to connect to Tor is O(log T) with high probability.
As my second goal, I address the problem of multi-party shuffling, where multiple parties, each holding an input, want to agree on a random permutation of their inputs while keeping the permutation secret. Multi-party shuffling is a primitive in many privacy-preserving applications such as anonymous communication and electronic voting. We describe an unconditionally-secure protocol for this problem that scales well to large networks; it requires each party to send (and compute) a polylogarithmic number of bits (and operations) while incurring a logarithmic round complexity. Our scheme is secure against traffic analysis and malicious faults from up to a third fraction of the parties.
Bio: Mahdi Zamani is a doctoral candidate in Computer Science at the University of New Mexico, Albuquerque. He is working with Jared Saia on topics in fault-tolerant distributed algorithms and randomized algorithms. In particular, he is interested in designing randomized algorithms for anonymous communication systems. Mahdi holds a master's degree in Computer Science from the University of New Mexico and a bachelor's degree in Computer Engineering from Tehran Polytechnic -- Read about Mahdi's current projects at: http://cs.unm.edu/~zamani