Security games with partial surveillance

Security games are used to deploy limited security resources. Much work on the topic assumes that attackers have the perfect knowledge of defenders’ strategies. However, it is not always the case in real life because an attacker may worry he will be caught if he observes defender’s strategy on all targets. To address the issue, this paper proposes a new game model in which the attacker just selects partial targets to observe according to his goal and observation cost. Moreover, our theoretical and experimental analyses show that our model reflects well the way that attackers make decisions, and in particular, the defender can gain significantly higher utility by considering the attacker can only conduct observations on partial targets.