Seminar: Application of Evolutionary Algorithms to Security Games

Globalization of threats to homeland security, such as international terrorism, smuggling of weapons or drugs, or large-scale thefts became one of the main challenges for security forces in the 21st century. In effect, new scientifically grounded methods for fighting organized crime and terrorist threats have been proposed and developed in recent years. One of the rapidly developing research areas are Security Games (SG), which model tactical security issues as games played between security forces and organized attackers. In this talk I will introduce our recently-proposed metaheuristic approach to approximation of the optimal defender’s strategy in general-sum sequential Stackelberg Security Games with imperfect information. The method (Evolutionary Approach to Security Games - EASG) employs Evolutionary Algorithms with specially designed chromosomes and genetic operators. Experimental evaluation indicates that EASG scales in time and memory better than state-of-the-art MILP (Mixed Integer-Linear Program) methods while providing optimal or close-to-optimal solutions in the vast majority of the cases. Except for competitive time and memory scalability, an additional asset of EASG is flexibility, as the method is generic and largely game-independent. Furthermore, EASG is anytime method, i.e. it can be terminated in any moment and still provide a reasonably good solution, which makes it particularly well suited for time critical SG applications. The talk will also cover several enhancements to the baseline EASG formulation, including an addition of memetic operations (local optimization procedures), the use of two competing populations (co-evolutionary approach), or neuro-evolutionary method in which the attacker’s decision-making model is approximated by a neural network. Application of the above-mentioned approaches to various security scenarios (warehouse patrolling, ferries protection, poaching prevention, cybersecurity) will be presented.

About the Speaker

Prof. J. Mańdziuk received M.Sc. (Honors) and Ph.D. in Applied Mathematics from the Warsaw University of Technology (WUT), Poland in 1989 and 1993, resp., and D.Sc. degree in Computer Science from the Polish Academy of Sciences in 2000. He is a full professor at the Faculty of Mathematics and Information Science, WUT. His research interests include application of Computational Intelligence and Artificial Intelligence methods to games, dynamic and bilevel optimization problems, and human-machine cooperation in problem solving. He is also interested in the development of general-purpose human-like learning and problem-solving methods. For more information please click here.
 

To arrange a personal meeting with Prof. J. Mańdziuk: please contact his host, Dr. Ami Moshaiov at moshaiov@tauex.tau.ac.il