GRIST Lab Develops Security Protocols Via Gaming

What if playing cops and robbers could inform how defense strategies are built for response to real attacks? That’s the thinking behind the GRIST project that is being developed at Rutgers School of Engineering’s Department of Industrial and Systems Engineering.

“Using game theory, the study of mathematical models of conflict and cooperation between intelligent decision makers, we can build a game where we would divide the decision makers as attackers and defenders, then use the game to better understand which infrastructures an attacker would go for,” said Melike Baykal-Gürsoy, director of the Laboratory for Stochastic Systems and the lead researcher on the project.

Along with ISE doctoral candidates Abdolmajid Yolmeh and Zhifan Xu, computer science students Mudassar Zaidi, Jennifer Borucki, Andrew Choliy, and the director of the Virtual Worlds, Rick Anderson, Baykal-Gursoy is creating a playable game for humans to compete against each other. Attackers simulate attacks on certain targets, while defenders figure out how to block them.

The project, which has received support from the National Science Foundation, creates an immersive virtual world that intelligently tracks how players behave and can learn from the activities performed by them. From this data, researchers can simulate how actual terrorists might behave in a given situation, helping law enforcement predict and plan responses for attacks. In addition to the game itself, Baykal-Gürsoy’s team intends to collect data by analyzing specific targets like transit facilities, banks and other public areas.

“This project is related to seeing some of the news about attacks. I noticed that crowds are important for some attackers, and you want to limit those attacks,” she said.

The team has already built a simplified version of the game on their website, which they use to collect data and demonstrate their concept. This current game is two-dimensional, but the more complete game will be three-dimensional.

“We demonstrated the 2-D game at Rutgers Day and the School of Engineering’s open house and collected the data,” Baykal-Gürsoy said. “We are now building a 3-D game, which will be more informative.”

The project has several distinct areas of focus unified under the central theme of using security game play data to gain insight into attacker models. One focus emphasizes what policies an attacker must follow to inflict the most damage and what policies a defender should have to limit the attackers success. Player behaviors are studied to analyze how actual attackers and defenders might react to different situations.

Another focus is on member behavior, but allows for differences in what each attacker or defender might know. Known as “private information,” this research assumes that the unique ideas and knowledge of different participants will form the basis for their attack or defense strategies, similar to real-world situations. The focus considers variable parameters as a factor, and will help defenders optimize their work while assuming they do not have all the available information. Still another looks at how players allocate technical resources to alter detection capabilities in addition to the defend-attack game.

Combining these different areas of research can result in a comprehensive game while simplifying the development process for each researcher involved. Along with undergraduate students from the Aresty Research Program, the doctoral candidates working under Baykal-Gürsoy will determine the most efficient and successful methods for creating a game that can simulate an attack. Data from the game can then be used to help model actual attacks from terrorists.

Preliminary results show that attackers are more likely to attack the first possible target, while defenders focus on high-value locations rather than all possible targets.

“With these in mind, we were able to come up with an algorithm that can improve the probability of the attacker being captured,” Baykal-Gürsoy said.

She hopes that her work will help law enforcement and other groups identify new strategies to deal with public safety and security.

“There are different types of attackers who operate differently,” she said. “We want to know, can you learn from the attackers?”

Interested in testing the game? Check out grist.rutgers.edu.