Andrew J. Collins

Probabilistic Risk Analysis and Terrorism Risk

Since the terrorist attacks of September 11, 2001, and the subsequent establishment of the U.S. Department of Homeland Security (DHS), considerable efforts have been made to estimate the risks of terrorism and the cost effectiveness of security policies to reduce these risks. DHS, industry, and the academic risk analysis communities have all invested heavily in the development of tools and approaches that can assist decisionmakers in effectively allocating limited resources across the vast array of potential investments that could mitigate risks from terrorism and other threats to the homeland. Decisionmakers demand models, analyses, and decision support that are useful for this task and based on the state of the art. Since terrorism risk analysis is new, no single method is likely to meet this challenge. In this article we explore a number of existing and potential approaches for terrorism risk analysis, focusing particularly on recent discussions regarding the applicability of probabilistic and decision analytic approaches to bioterrorism risks and the Bioterrorism Risk Assessment methodology used by the DHS and criticized by the National Academies and others.

Comparing Reinforcement Learning Approaches for Solving Game Theoretic Models: A Dynamic Airline Pricing Game Example

Games can be easy to construct but difficult to solve due to current methods available for finding the Nash Equilibrium. This issue is one of many that face modern game theorists and those analysts that need to model situations with multiple decision-makers. This paper explores the use of reinforcement learning, a standard artificial intelligence technique, as a means to solve a simple dynamic airline pricing game. Three different reinforcement learning approaches are compared: SARSA, Q-learning and Monte Carlo Learning. The pricing game solution is surprisingly sophisticated given the games simplicity and this sophistication is reflected in the learning results. The paper also discusses extra analytical benefit obtained from applying reinforcement learning to these types of problems.

Mortgage Modification and the Decision to Strategically Default: A Game Theoretic Approach

While numerous and varied opinions abound, there remains much confusion as to why relatively few mortgages are modified at a time when the demand to modify is historically high. To better understand this complex issue, we build a game theoretic model to quantify a number of economic incentives and costs surrounding critical dimensions of the lenders decision to modify a loan and the borrowers decision to strategically default in an attempt to encourage such a modification. We mathematically demonstrate that it is rarely economically rational for lenders to modify loans. For the borrower, we find that their negative equity position, growth rate in home prices, and the probability the lender will exercise its legal right to recourse represent the top three strategic default determinants.

Conceptualizing intragroup and intergroup dynamics within a controlled crowd evacuation

Social dynamics play a critical role in successful pedestrian evacuations. Crowd modeling research has made progress in capturing the way individual and group dynamics affect evacuations; however, few studies have simultaneously examined how individuals and groups interact with one another during egress. To address this gap, the researchers present a conceptual agent-based model (ABM) designed to study the ways in which autonomous, heterogeneous, decision-making individuals negotiate intragroup and intergroup behavior while exiting a large venue. A key feature of this proposed model is the examination of the dynamics among and between various groupings, where heterogeneity at the individual level dynamically affects group behavior and subsequently group/group interactions. ABM provides a means of representing the important social factors that affect decision making among diverse social groups. Expanding on the 2013 work of Vizzari et al., the researchers focus specifically on social factors and decision making at the individual/group and group/group levels to more realistically portray dynamic crowd systems during a pedestrian evacuation. By developing a model with individual, intragroup, and intergroup interactions, the ABM provides a more representative approximation of real-world crowd egress. The simulation will enable more informed planning by disaster managers, emergency planners, and other decision makers. This pedestrian behavioral concept is one piece of a larger simulation model. Future research will build toward an integrated model capturing decision-making interactions between pedestrians and vehicles that affect evacuation outcomes.

Generic Incident Model for Investigating Traffic Incident Impacts on Evacuation Times in Large-Scale Emergencies

Traffic incidents cause a ripple effect of reduced travel speeds, lane changes, and the pursuit of alternative routes that results in gridlock on the immediately affected and surrounding roadways. The disruptions caused by the secondary effects significantly degrade travel time reliability, which is of great concern to the emergency planners who manage evacuations. Outcomes forecast by a generic incident model embedded in a microscopic evacuation simulation, the Real-Time Evacuation Planning Model (RtePM), were examined to quantify the change in time required for an emergency evacuation that results from traffic incidents. The incident model considered vehicle miles traveled on each individual segment of the studied road network model. The two scenarios considered for this investigation were evacuations of (a) Washington, D.C., after a simulated terrorist attack and (b) Virginia Beach, Virginia, in response to a simulated hurricane. These results could help the emergency planning community understand and investigate the impact of traffic incidents during an evacuation.

Agent-based modeling and strategic group formation: a refugee case study

Refugee flight presents a logistics problem for humanitarian aid workers anticipating ebbs and flows of arrivals. These migrations include travel over long distances with little advanced coordination and damaged social networks. The model presented here is based on these two fundamental premises: long distance and strategic en-route coordination. Assuming that over large distances, refugees attempt to construct groups that provide assistance or security as they navigate toward safety, the agent-based model incorporates cooperative game theory to investigate the impact of group formations on egress times. The modeled refugees make decisions based on individual utility functions informed by two factors, speed of the group and group size. Since groups accommodate slower members, they may reform as refugees choose their best available strategies to reach safety. The results indicate a tipping point in average group size as the slowest group members have more of impact in the utility function of the agents.

Panic That Spreads Sociobehavioral Contagion in Pedestrian Evacuations

Crowds are a part of everyday public life, from stadiums and arenas to school hallways. Occasionally, pushing within the crowd spontaneously escalates to crushing behavior, resulting in injuries and even death. The rarity and unpredictability of these incidents provides few options to collect data for research on the prediction and prevention of hazardous emergent behaviors in crowds. This study takes a close look at the way states of agitation, such as panic, can spread through crowds. Group composition—mainly family groups composed of members with differing mobility levels—plays an important role in the spread of agitation through the crowd, ultimately affecting the exit density and evacuation clearance time of a simulated venue. This study used an agent-based model of pedestrian movement during the egress of a hypothetical room and adopted an emotional, cognitive, and social framework to explore the transference and dissipation of agitation through a crowd. The preliminary results reveal that average group size in a crowd is a primary contributor to the exit density and evacuation clearance time. The study provides the groundwork on which to build more elaborate models that incorporate sociobehavioral aspects to simulate human movement during panic situations and account for the potential for dangerous behavior to emerge in crowds.

An improvement selection methodology for key performance indicators

Key performance indicators (KPIs) are critical measures for determining the health of a manufacturing plant in relationship to the plant’s goals. In today’s competitive environment, manufacturers cannot be careless about their business; in fact, they must ensure that their KPIs are effective and use them to make improvements when necessary. This paper describes a method for suggesting improvements to a manufacturer’s KPIs, based on the results achieved from a workshop to score the KPI on a number of predefined criteria. The approach uses a prospect theory approach to weight the scoring. Different problem formulations were derived that allow for both recommendations for improvements and the recommendations for disinvestments to over-performing KPIs. The authors applied the developed approach to two workshop outputs, each from independent manufacturers, and the results highlighted the significant difference between the two manufacturers in terms of improvement priorities and KPI assessment. The optimal improvement suggestions were compared to those found through a fast heuristic. It was determined that given the underlying assumptions of the approach that the heuristic solutions were just as adequate as the optimal ones.

Popularity or proclivity?: revisiting agent heterogeneity in network formation

Agent-based modeling (ABM) approach is used to reassess the Barabasi-Albert (BA) model, the classical algorithm used to describe the emergent mechanism of scale-free networks. This approach allows for the incorporation of agent heterogeneity which is rarely considered in BA model and its extended models. The authors argue that, in social networks, peoples intention to connect is not only affected by popularity, but also strongly affected by the extent of similarity. The authors propose that in forming social networks, agents are constantly balancing between instrumental and intrinsic preferences. The proposed model allows for varying the weighting of instrumental and intrinsic preferences on the agents attachment choices. The authors also find that changing preferences of individuals can lead to significant deviations from power-law degree distribution. Given the importance of intrinsic consideration in social networking, the findings emerged from this study is conducive to future studies of social networks.

Applying reinforcement learning to an insurgency Agent-based Simulation:

A requirement of an Agent-based Simulation (ABS) is that the agents must be able to adapt to their environment. Many ABSs achieve this adaption through simple threshold equations due to the complexity of incorporating more sophisticated approaches. Threshold equations are when an agent behavior changes because a numeric property of the agent goes above or below a certain threshold value. Threshold equations do not guarantee that the agents will learn what is best for them. Reinforcement learning is an artificial intelligence approach that has been extensively applied to multi-agent systems but there is very little in the literature on its application to ABS. Reinforcement learning has previously been applied to discrete-event simulations with promising results; thus, reinforcement learning is a good candidate for use within an Agent-based Modeling and Simulation (ABMS) environment. This paper uses an established insurgency case study to show some of the consequences of applying reinforcement learning to ABMS, for example, determining whether any actual learning has occurred. The case study was developed using the Repast Simphony software package.