I. Introduction
Our systems today are vulnerable to Sybil attacks, in which an attacker injects multiple fake accounts into the system [1] . Recently, the increasing popularity of online social networks has made them attractive targets for Sybil attacks. It is estimated that tens of millions of Sybil accounts exist in popular social networks such as Facebook and Twitter [2] , [3] . The attacker can leverage Sybil accounts to disrupt democratic election and influence financial market via spreading fake news [4] , [5] , as well as compromise system security and privacy via propagating social malware, disseminating scams, and learning users’ private data [2] , [3] , [6] – [10] .