Project Outcomes Report

The primary aim of this project was to address the gaps in current understanding of lone actor attacks to guide the development of new innovative defensive policies. In this award, we analyzed interactions between attackers and defenders across different scenarios and uncertainties, and suggested defense strategies through game-theoretic framework and empirical data. The outcomes are categorized under the following sections: 

1. Real-world data collection and behavior analysis:

We gathered and collected lone-actor terrorism data and made it available through a publicly available site. Our preliminary analysis of data revealed that while pre-9/11 lone actors mostly radicalized in their own environment, post-9/11 lone actors are more diverse.  Moreover, our analysis also made the scarcity of incident scene data more evident. 

We contrasted, as a proxy, the 2D-game data generated at the Game Research for Infrastructure SecuriTy (GRIST) Lab and real world lone-actor terrorism data and identified five main types of attackers regarding their target selection and attack preparation schemes, out of which ``maximum damagers'' are the most critical. In addition, we used Association Rule Mining that demonstrated the implications among radicalization and attack preparation behaviors. These implications produced temporal behavioral chains on the pathway to violence. The study also points out the unavailability of the incident-scene data. 

2. Serious game development and data collection for incident-scene behavior:

In response to the incident-scene data scarcity, serious games serve as a suitable surrogate for data generation. To this end, we designed the "Paint Fever" game (http://grist-lab-ise.rutgers.edu/games.html) to provide a platform for analyzing attacker behavior. The game simulates a critical infrastructure environment featuring individuals and a disguised attacker, with defenders patrolling and locating the attacker. This game provided valuable publicly available data to analyze and understand attackers' incident-scene behavior.

Data from interactive games revealed that the attackers' micro-trajectories significantly expose them during incident scenes. Attackers' behaviors at the incident scene include instinctively reverting upon an encounter with a defender or attempting to blend seamlessly with the surrounding crowd. However, these micro-trajectories are mostly instinctive and as such can be detectable. 

3. Developing defense policies against various attack scenarios:

We considered various security problems, for example, search games, and defensive resource allocation scenarios. In these situations, the ultimate goal of the defender(s) is to promptly locate the attacker and the attack weapon in order to prevent any harm to the public. We used game-theoretic and optimization arguments to find the defender's optimum strategy for these problems. For specific cases, whenever it is possible, we proved that the defender's optimum strategy is unique and can be simply formulated. In the general cases, we provided algorithms to computationally solve for the optimum strategy. We showed that these algorithms possess properties that demonstrate their operational effectiveness and efficiency.

A more extensive defense policy includes overarching protection. An unrealistic assumption is that only one type of protection protects all targets. Overarching protection necessitates a cost-effective combination and allocation of various measures against multiple attacks or disasters. In this award, we showed that a country-wide problem can be decomposed into smaller city-level subproblems and solved optimally at each scale, including both static and dynamic protection resources. 

4. Extension to cyber-security attacks:

The attackers do not always aim to maximize the damage to the public; in some cases, they prefer to gather intelligence, or to disrupt the communication between defenders in order to create chaos. If the attacker's type is unknown, we showed that the best defense strategy involves an optimal mix of the strategies against each type of attackers separately.

We considered the problem of increasing the communication secrecy throughout a network by applying jamming to eavesdropping attackers. The results of this award indicated that the optimum policy maintains an effective balance between the jamming strength and clear communication among viable users. The conservative approach in such problems assumes an ominous eavesdropper attacking all communication channels simultaneously. However, We showed that strategic attacks distribute the risk to different channels. Against such attacks, we proved that the optimal policy is unique and follows a water-filling scheme. 

5. Publications and Dissertations:

The National Science Foundation funding provided support for six doctoral students, and one undergraduate student, with four doctoral students receiving partial funding as part of their participation and three having completed their dissertations. The results of this research had been disseminated as four journal articles and five refereed conference proceedings, and eight conference presentations. 

Last Modified: 11/02/2023
Modified by: Melike Baykal-Gursoy