The Power of Information in Security Games

People

Haifeng Xu
Milind Tambe
Shaddin Dughmi

Acknowledgement


Motivation

Asymmetric information can help improve security in Security Games.
Research in traditional Stackelberg security games allocates physical resources according to the Strong Stackelberg Equilibrium (SSE) mixed strategy. This model answer standardizes the modeling and analysis of security games, however unfortunately, on the other hand side, also clouds our vision for a further understanding of defense. What else can we exploit to build or increase the defense besides physical resources? This project studies another important, yet largely unexplored, defensive resource - information.
Information - i.e., who knows what knowledge regarding a game - has a profound influence on the equilibrium outcome of the game. From the informational perspective, defense is all about shaping the attacker's belief regarding the protection of targets, and randomly allocating physical resources is just one way to achieve this. The attacker's belief is also largely affected by the information available to him, such as payoff structures, effectiveness of physical resources, vulnerability of targets, defense deployments, etc. Crucially, the defender usually has more knowledge regarding these aspects than the attacker. The central question we aim to answer in this project is, can the defender make use of such knowledge to increase the defensive effects, and if so, how she can do it optimally?

Information as a Means of Defense

In most security domains, the defender has more information regarding the game than the attacker in different aspects. For example, in ferry protection domains, the defender knows which ferries have more people; in port protection domains, the defender knows which ports are more vulnerable; in domains with various patrolling resources, the defender knows which resources are more effective. The attacker may have only partial or even no information regarding these aspects. This raises a general question: can this information be strategically used for defense? Our paper in AAAI 2015 answers this question in the affirmative in a natural two-stage game model, and also provides an efficient approach to do this optimally. To best of our knowledge, this is the first published work that starts to address this question formally. The information we exploit is the defender's private knowledge about the instantiated deployment at each round, which is not known by the attacker. Surprisingly, our approach is to reveal to the attacker information about the instantiated deployment, but strategically. We refer the readers to the paper for further information. Our ongoing work begins to study more general settings and explore the strategic usage of other types of information.

Publications

Haifeng Xu, Zinovi Rabinovich, Shaddin Dughmi, Milind Tambe. AAAI 2015.