Research in privacy-preserving data publishing has revealed the necessity of accounting for an adversary's background knowledge when reasoning about the protection afforded by various anonymization schemes. Most existing work models the background knowledge of one individual adversary or privacy attacker, or makes a worstcase assumption that attackers will act as one: colluding through sharing of background information. We propose a framework for modeling multiple attackers with heterogeneous background knowledge, supporting analysis of their strategic incentives for sharing information prior to attack. The framework posits a decentralized mechanism by which agents decide whether and how much information to share, and defines a normal-form game representing their strategic choice setting. Through a simple example, we show that the efficacy of database generalization operations depends on the information-sharing strategies adopted by the attackers. Through analysis of the underlyi...
Quang Duong, Kristen LeFevre, Michael P. Wellman