Playing Games for Airport Security
Earlier this month, it was the ten-year anniversary of the terrorist attacks of September 11th. Since then, what measures have airports taken to ensure safer travel? The major issue with airport security until recently was the limited security resources (e.g., canine patrols at airport gates, checkpoints on incoming roads), which prevented complete security coverage at all times. In other words, terrorists have had the advantage of exploiting patterns in police patrolling and monitoring.
Los Angeles International Airport (LAX) is the fifth busiest airport in the United States, serving between sixty to seventy million passengers per year. Researchers at the University of Southern California (USC) led by Computer Science Professor Milind Tambe have developed software that transforms the patrolling problem at LAX into a Bayesian-Stackelberg game, which allows police officers to “weigh their different actions in randomization, as well as uncertainty in adversary types.” The software is hence called Agent Modeling for Security Planning Using Randomization, or ARMOR for short, and has been employed by the LAX police force since 2007.
The game is far more complicated than those discussed in class; nevertheless, there are several defining characteristics. Firstly, it is a sequential or Stackelberg game, which means that the two “players” — the agent and the terrorist — do not act simultaneously. Instead, the airport acts first in setting up defense systems, daily. The terrorist then optimizes his or her own reward based on the action chosen by the agent. Secondly, the game is termed Bayesian because there is incomplete information about the terrorists, since each advisary has his or her own probability of causing terror. Furthermore, a particular advisary may prefer one form of terror over another, etc.
The Bayesian-Stackelberg game is solved using the Decomposed Optimal Bayesian Stackelberg Solver (DOBSS), which uses the fact that terrorists act independently of each other. The scope of DOBSS is beyond the scope of this post. In any case, the inputs in DOBSS provided by the police are possible terrorist targets and their relative importance, and the output is the “optimal mixed strategy for the police officer to commit to, … given that the agent may not be aware of the terrorist’s plans in advance.” The police can use this information to then decide when and where to place checkpoints on roads and canine units in terminals.
Next time I travel through security in LAX, I’ll be sure to notice the randomized strategies employed by the police force, generated from the ARMOR program to catch terrorists. While the application of game theory to the issue of security is indeed very interesting, I’m still somewhat skeptical: can terrorists actually be assumed to be rational, as required in this Bayesian-Stackelberg game? Also, practically speaking, how much expert input is needed to operate the software?
Source: http://teamcore.usc.edu/praveen/papers/jamesaimag08.pdf
I was curious if you ever thought of changing the page layout of your blog? Its very well written; I love what youve got to say. But maybe you could a little more in the way of content so people could connect with it better. Youve got an awful lot of text for only having one or 2 pictures. Maybe you could space it out better?