Security is a critical societal challenge.We focus on urban security: the problem of preventing urban crimes. The Stackelberg Security Game (SSG) was proposed to model highly strategic and capable adversaries who conduct careful surveillance and plan attacks, and has become an important computational framework for allocating security resources against such adversaries.
While there are such highly capable adversaries in the urban security domain, they likely comprise only a small portion of the overall set of adversaries. Instead, the majority of adversaries in urban security are criminals who conduct little planning or surveillance before “attacking”. These adversaries capitalize on local opportunities and react to real-time information. Unfortunately, SSG is ill-suited to model such criminals, as it attributes significant planning and little execution flexibility to adversaries.
Inspired by modern criminological theory, we introduce the Opportunistic Security Game (OSG), a new computational framework for generating defender strategies to mitigate opportunistic criminals.
The OSG model of opportunistic criminals has three major novelties compared to SSG adversaries: (i) criminals exhibit Quantal Biased Random Movement, a stochastic pattern of movement to search for crime opportunities that contrasts with SSG adversaries, who are modeled as committed to a single fixed plan or target; (ii) criminals react to real-time information about defenders, flexibly altering plans during execution, a behavior that is supported by findings in criminology literature; (iii) criminals display anchoring bias, modeling their limited surveillance of the defender’s strategy.
Outline of OSG
Human experiments on OSG
Applying Machine learning technique in OSG
To learn the OSG, we represent the criminal behavior and the interaction with the patrol officers as parameters of a Dynamic Bayesian Network (DBN), enabling application of different algorithms such as EM to learn the parameters.