CHISIM: AN AGENT-BASED SIMULATION MODEL OF SOCIAL INTERACTIONS IN A LARGE URBAN AREA

Full text

Proceedings of the 2018 Winter Simulation Conference
M. Rabe, A.A. Juan, N. Mustafee, A. Skoogh, S. Jain, and B. Johansson, eds.
CHISIM: AN AGENT-BASED SIMULATION MODEL
OF SOCIAL INTERACTIONS IN A LARGE URBAN AREA
Charles M. Macal
Nicholson T. Collier
Jonathan Ozik
Eric R. Tatara
John T. Murphy
Decision and Infrastructure Sciences
Argonne National Laboratory
9700 S. Cass Ave.
Argonne, IL 60439 USA
ABSTRACT
Cities are complex, dynamic, evolving adaptive systems comprised of people as well as interconnected
physical infrastructure. Simulation modeling can help us understand and shape the evolution of our cities.
In this paper, we describe an agent-based simulation modeling framework applied to Chicago, called
chiSIM (for the Chicago Social Interaction Model). Each person residing in Chicago is represented as an
agent in chiSIM; all places where people can be located in Chicago also are represented. The model
simulates the movements of people between locations on an hourly basis during the course of a typical day.
Co-located agents engage in various kinds of social interactions, such as exchanging information, engaging
in business transactions, or simply sharing physical proximity. We discuss technical approaches to large-
scale urban modeling including development of synthetic populations, efficiency gains through distributed
processing, logging and analysis of simulation results, and visualization.
1 INTRODUCTION
“A city is first and foremost a social reactor. It works like a star, attracting people and accelerating
social interactions” (Bettencourt 2013).
Cities are complex systems whose functioning depends upon many social, economic, and
environmental factors (Bettencourt 2013) . Many cities, called megacities, around the world are growing
exponentially in size and complexity. One hundred fifty years of rapid global urbanization have taken the
global urban population from 14% (in 1990) to 50% (in 2008) and projected to be 70% by 2050 (Population
Reference Bureau 2013). China alone is projected to have 221 cities with 1 million or more people by 2030,
with the addition of 400 million city dwellers; between now and 2020 (Dobbs 2010), the Guangdong
province will invest $229 billion in transport infrastructure to create a single 50 million person city.
There is a growing realization that cities are sources of potential opportunity as well as instability. With
this realization has come the need for methods and tools that can help us understand the processes by which
cities will grow and transform themselves in the coming years. Simulation has a lot to offer in understanding
the cities of the future, and helping to plan for sustainable growth Many analytical and modeling techniques
are finding value in developing an understanding of urban processes (Johnson et al. 2016). Computer
simulation models can capture the dynamic processes operating within the city environment and offer a
promising approach to addressing these and others questions. Many models have been developed for cities
that focus on the lifeline physical infrastructures: transportation, energy, water, communications, and
others; typically, these models ignore the interconnectivity between infrastructures and the people that
810978-1-5386-6572-5/18/$31.00 ©2018 IEEE

Macal, Collier, Ozik, Tatara, and Murphy
utilize the services provided by them. An important challenge is to include people and social processes into
models of cities. This paper describes a model applicable to large urban areas that includes people--their
behaviors, and social interactions--as central elements in shaping a city’s future.
chiSIM, the Chicago Social Interaction Model, is an agent-based model framework of people and places
in Chicago along with the daily activities in which they engage. Agent-based simulation, a relatively new
approach to modeling populations of heterogeneous, interacting, adaptive agents (Macal 2016) at an
individual, granular level of detail, is the technical simulation approach taken in chiSIM. chiSIM models
the behaviors and social interactions of all Chicago residents. Agents consist of the population of all the
residents of Chicago represented at the individual level; places consist of geo-located parcels in the city,
such as households, schools, workplaces, hospitals, and general quarters, such as nursing homes,
dormitories, jails, etc. During the course of a simulated day, agents move from place-to-place, hour-by-
hour, engaging in social activities and interactions with co-located agents.
The chiSIM framework has many potential uses such as forecasting the need for transportation, energy
and other infrastructure services; the spread of infectious diseases; the spread of information; the adoption
of new technologies; the effectiveness of social programs; and many other uses. To date, chiSIM has been
used to understand what measures could be used to mitigate the spread of community-associated MRSA
(Methicillin-resistant Staphylococcus aureus), a skin and soft tissue infection caused by Staphylococcus
aureus bacteria (Macal et al. 2014), and a community healthcare information program (Kaligotla et al.
2018). Work in progress includes modeling the transportation needs of the population on the infrastructure
including modeling charging station location (electric vehicles) and fueling station location (hydrogen-
powered vehicles), HIV, Hepatitis C and other infectious disease transmission in high-risk communities,
and forecasting the individual decisions on residential energy consumption and the consumer adoption of
residential rooftop solar energy technologies.
2 BACKGROUND
Modeling cities and their populations has a long history, originating with the invention and rising popularity
of simulation as a field, chronicled by Batty (2008), who also notes that urban modeling was driven in part
by the availability and increasing power of computers, though these early efforts often exceeded the
computing power available. Early urban models, in the 1950’s and 1960’s onward, were systems dynamics
approaches (Forrester 1969). Cellular automata (CA) then enjoyed a heyday in the late 1980’s and 1990’s
and eclipsed SD models at least in terms of the extent of the literature (Batty 1989). Most of the modeling
effort uses CA to examine urban growth or land use change, and the systems dynamics approach are broadly
economic (industry output, labor pools, etc.).
Other approaches like 3-D modeling and addressing urban research through “big data” approaches are
becoming more prominent. Cities have been studied from the standpoint of complexity. One research
program focuses on general or universal principles characterizing cities and their macroscopic behaviors.
For example, scaling laws have been derived to predict the average social, spatial, and infrastructural
properties of cities as a set of scaling relations. There are also system- or infrastructure specific urban
simulations focusing fir example on traffic, pedestrian movement, disease spread, disaster response and
evacuation, etc.
Agent-based models (ABM) were introduced around the early 2000’s. ABMs have been applied to a
very broad set of applications including: traffic and transportation, pedestrian movement, social contact
networks, crime, segregation, disease, natural resource management and sustainability, economics and real
estate, disasters, and others. We use an agent- based modeling approach in chiSIM because the agent
approach can best represent the diversity of an entire population in terms of the characteristics and behaviors
of individuals. Few if any urban models exist at the large scale of the chiSIM, in terms of the number of
individuals and behaviors considered at the fine-grained level of geo-spatial detail for such a large
geographical area.
811

Macal, Collier, Ozik, Tatara, and Murphy
3 SIMULATING PEOPLE, PLACES, AND ACTIVITIES
Assembling the large amount of information from a variety of heterogeneous data sources to support a city-
scale ABM is a challenging technical and logistical problem. An agent-based city model begins with a
synthetic population of individuals (people) who are represented as agents in the model (Figure 1). There
is no data on actual, specific individuals in the model but characterizations of individuals that are accurate
when aggregated. Synthetic populations with baseline socio-demographic data are derived from combined
U.S. Census files that are available from a growing number of sources. chiSIM uses baseline synthetic
population data such as those developed by Wheaton et al. (2009). The socio-demographic attributes of the
synthetic population match that of the actual population for Chicago in the aggregate for the Census years
of 2000 and 2010. Each agent has a baseline set of socio-demographic characteristics (e.g., race/ethnicity,
age, gender, educational attainment, income).
Figure 1: chiSIM agents.
An agent-based city model also includes a synthetic population of places (Figure 2). All places are
characterized by place type, including households, schools, hospitals and workplaces, and have locations.
The synthetic population assigns agents to households, workplaces and schools (for those of school age).
Places are categorized as having different types of activities that may occur there.
We combine several publicly available national data sources to model activities of the synthetic agent
population. Each agent has a daily activity profile that determines what times throughout the day he or she
occupies each location. Activity profiles are empirically based on 24-hour time diaries collected as part of
the U.S. Bureau of Labor Statistics’ annual American Time Use Survey (ATUS) for individuals aged 15
years and older and from the Panel Study of Income Dynamics (PSID) for children younger than 15 years.
Both are nationally representative samples and collect diary data on randomly assigned days. The diary
records each activity during the 24-hour period (start/stop times, location and others present). Two profiles
(one weekday and one weekend) from respondents living in metropolitan areas are assigned to each agent
Saturday - Sunday
Hourly Itinerary
by day
12:00 am
sleep
8:00 am work
5:30 pm home
6:30 pm grocery
shop
8:00 pm socialize
Monday - Friday
Household
ID
Type
Latitude
Longitude
School
Virtual
Agent
. . . Behavior
ID
Place Type
Risk
. . .
Behavior
. . .
Behavior
. . .
Synthetic Population
Activity Data
ID
Type
Latitude
Longitude
ID
Age
Gender
Income
Race/Ethnicity
Household
Workplace
School
…
Travelers
into region
Hospital ID
Type
Latitude
Longitude
Travelers out
of region
812

Macal, Collier, Ozik, Tatara, and Murphy
in the model. This is done by stochastically matching each agent with an ATUS or PSID respondent who
is either identical or similar with respect to socio-demographic characteristics.
Figure 2: chiSIM map of places in Chicago.
Basic agent behaviors, which can be gleaned from activity data, affect activity choice (what activities
do agents engage in at any time), and movement (when and where do agents change their location to engage
in a new activity). More specialized behaviors are application specific, including communication (what
specific information is exchanged in social interactions and with whom) and actions in unusual or specific
circumstances. For example, in the community health information study, a sampling of healthcare recipients
were surveyed on who and how often they pass on information from healthcare providers regarding
healthcare services available in their community. In a pandemics simulation, a behavioral ontology was
developed from the literature that enumerated actions people take in the event of a pandemic and the
813

Macal, Collier, Ozik, Tatara, and Murphy
circumstances under which those actions were taken. The behavioral ontology then was implemented in the
agent-based model to drive agent behaviors.
4 CHISIM FRAMEWORK OVERVIEW
The chiSIM frameworks enables building hybrid time-stepped and discrete-event, stochastic simulations.
The prototypical chiSIM model moves 3 million individual agents to and from 1.2 million places in Chicago
on an hourly basis. The simulation is initialized at hour zero with agents located across Chicago. The
Generalized Mobility/Activity Model advances agent locations appropriate for the hour of the day, moving
agents between their own households, schools, workplaces, other households, etc. Specialized
mobility/activity models are developed separately to model the circulation of agents from the community
into and out of special places that are not included in the general activity profile data, since that is developed
for general populations. Once activity state changes are recorded, or logged, the hour is incremented and
the simulation processes are repeated until the simulation reaches the end time, at which point yearly and
hourly summary reports are produced. chiSIM has simulated agent movement and activities by hour over
10 years. For example, to explore possible explanations for the ongoing MRSA epidemic in Chicago, the
model was simulated for the period 2001-2010 and was able to reproduce the empirical buildup that was
observed. For the 10-year hurly simulation (It was estimated that over 2 trillion total people-to-people
contacts were simulated over the 87,600 hourly time steps in the simulation.)
A large-scale agent-based model such as chiSIM presents challenges for computational resources. For
example, the MRSA ABM originally ran for 60 hours on a single compute node. Long run times for
simulations may severely limit the usefulness of a model for policy analysis. Collier et al. (2015) explored
the benefits of distributing a model across multiple computing nodes. They ran a series of experiments that
exploited the location/movement network structure by partitioning locations into separate computing
processes based on the expected people flow between the locations. Compute node thread parallelization
using OpenMP and distributed parallelization across multiple processes using MPI reduced the run time to
4 hours utilizing 128 compute nodes. Other distributed computing strategies promise to yield additional
benefits.
One of the challenges of working with a complex agent-based model is to record the “history” that the
model creates. The history records all the agent events by location and time over the course of the
simulation. Information on every agent’s state and location, and all the other agents that an agent is having
contact with at every moment, is known in the simulation. The simulation can be critical source of
information that is otherwise unobserved or even unobservable. For example, in the MRSA model, two
disease states are critical for transmission of the disease: the infected state, when an open skin or soft tissue
infection is readily observable, and the colonized stated, which is asymptomatic. A colonized individual
can transmit the disease but does not know they are contagious and can pass on the disease. The simulation
produced a reasonable recreation of the profile of the number of infected individuals (observable, and based
on empirical data) with MRSA; and the model was able to infer the correlated number of individuals that
were also colonized, which was not observed and for which there no empirical data exists.
The amount of data produced by the agent-based model of millions of interacting individuals moving
among millions of locations can be enormous. For example, Figure 3 shows the occupancy patterns of
agents by location at specific times. Figure 4 shows movement patterns for selected agents between
consecutive hours of the day. Figure 5 shows simulated disease transmission as a result of personal contact
between individuals. Because the log files can be enormous in size, collecting all the agent event data over
the simulation period, special techniques are required to record and analyze such large-scale datasets (Tatara
et al. 2017). Figure 6 shows the results of such an analysis, a co-location network for all agents in the model
who have two degrees of separation. This information is useful in infectious disease applications for tracing
agent contact patterns back to the original source, effectively identifying “patient zero.”
814

Macal, Collier, Ozik, Tatara, and Murphy
Figure 3: City occupancy by agents.
Figure 4: Agent movement patterns.
815

Macal, Collier, Ozik, Tatara, and Murphy
Figure 5: Agent disease transmission tracing in chiSIM.
Figure 6: Sample co-location network for randomly sampled individuals and all individuals within two
degrees of separation. Lines represent collocation between person nodes. Node color indicates vertex
degree, with darker nodes having a greater degree than lighter nodes.
816

Macal, Collier, Ozik, Tatara, and Murphy
5 TECHNICAL CHALLENGES IN SIMULATING A CITY
One of the primary challenges in implementing a large city-size simulation is achieving sufficient
computational performance such that model run time and memory use does not limit the utility of the model
itself. Implemented in C++, using the Repast for High Performance Computing (Repast HPC) (Collier and
North 2013), an agent-based model framework for implementing distributed agent-based models on high
performance distributed-memory computing platforms, chiSIM achieves such performance by distributing
a simulation across multiple processes.
As noted above, chiSIM began as a generalization of the model of community associated methicillin-
resistant Staphylococcus aureus (CA-MRSA) (Macal et al. 2014), mentioned above in section 3. The CA-
MRSA model was a non-distributed model in which all the model components, including all the people and
places required to simulate the city of Chicago, ran on a single computational process and thus often pushed
both memory bounds of individual computing nodes and run time limits of shared computing resources.
chiSIM retains and generalizes the social interaction dynamics of the CA-MRSA model and allows models
implemented using chiSIM to be distributed across multiple processes. Places are created on a single
process and remain there. Persons move between the processes according to their activity profiles, which
are generated from a problem-specific activity model describing how persons move about the community.
When a person selects a next place to move to, the person may stay on its current process or it may
have to move to another process if its next place is not on the current process. This cross-process distribution
ameliorates both run time and memory issues. By having fewer people and places on each process, and
allowing for processes to run in parallel, the time to iterate through each person’s and place’s behavior is
reduced. N number of people can be split in to N / number of processes sets of people, each of which runs
in parallel. Similarly, fewer people and places on a process also reduces an individual process’ memory
requirements.
However, parallelizing and distributing code across multiple processes is not without its own
performance pitfalls. Transferring data, that is, the internal state of people in chiSIM, between processes
can be computationally expensive. In a previous paper, we describe how we used the Metis graph
partitioning application to assign places to processes to minimize this computationally expensive cross-
process movement of persons and to balance the number of persons on each process (Collier et al. 2015).
In addition, chiSIM also provides the ability to cache any constant person state, given sufficient memory,
lessening the amount of data transferred between processes. When a person is moved between processes
only the dynamic state needs to be moved, any constant state can be retrieved from the cache when the
person is unpacked by the retrieving process.
As a software framework, chiSIM provides both the structure and an application programming interface
(API) for implementing, and more significantly distributing, a city-sized simulation across multiple
processes. The API allows the user to easily implement cross-process person movement as well as the code
that drives such movement, such as via activity schedules. In addition, it also provides flexible scheduling
and data collection (via Repast HPC), utility code for reading input files, working with model parameters
and other simulation implementation tasks, as well as providing some example simulations. A user of
chiSIM needs to provide the synthetic population, places and activity schedules, and implement any model-
specific behavior for persons, places, and activities, using chiSIM provided base classes and utility code.
The software stack for a typical chiSIM-based model is shown in Figure 7. For example, given a city-
sized epidemiological simulation in which people spread a disease through co-location with other people,
the user needs to provide as input data the synthetic population, the places in which people can be co-
located, and activity schedules that describe when the people move to those places. The model specific
behavior that needs to be implemented are such things as disease state transitions within each person, the
disease transmission between co-located people, and the selection of a destination place given an activity
schedule. The latter can be as simple as a direct mapping between time and place (e.g., at nine o’clock go
to work) or a more complex dynamic decision that may depend on additional person and simulation state.
817

Macal, Collier, Ozik, Tatara, and Murphy
These model-specific components are built on chiSIM provided code, and integrated into the chiSIM
provided structure to create the working model. The chiSIM framework is free and open source, and is
available at github.com/Repast/chiSIM. The chiSIM repository contains detailed instructions for building
and using the framework, along with example chiSIM-based simulations.
Figure 7: The software stack for a chiSIM-based model.
A chiSIM model enables in silico experimentation of city-scale phenomena. However, due to the
stochastic processes within a typical model and its likely large parameter space, tasks such as model
calibration and validation, characterization of the model input parameter space, propagating of input data
uncertainties, and the computational discovery of effective interventions, tasks needed for both creating a
trusted model and utilizing it, require the ability to run large ensembles of model instances based on
complex and iterative logic, what we term model exploration (ME). Examples of ME algorithms include
Approximate Bayesian Computing (Beaumont 2010), Active Learning (Settles 2012), and
genetic/evolutionary algorithms (Holland 1992). Given the potentially significant computation involved in
running a single distributed chiSIM model, coordinating an algorithm that dynamically generates
collections of models based on prior iterations is a non-trivial endeavor. To address the need for running
large-scale ME experiments, we have developed the Extreme-scale Model Exploration with Swift
(EMEWS) framework (Ozik et al. 2016) to enable the creation of automated, iterative HPC workflows.
EMEWS, built on top of the Swift/T parallel scripting language (Wozniak et al. 2013), enables the
framework user to directly plug in a distributed MPI-based model and an existing ME algorithm
implementing potentially complex iterating logic to create a large-scale HPC workflow. ME algorithms can
be expressed in popular data analytics languages such as Python and R. Despite its flexibility an EMEWS
workflow is highly performant and scalable to the largest HPC resources. Large EMEWS workflows of
chiSIM models have been run on a variety of computing resources, including the Mira BlueGene/Q
supercomputer, Cray XE6 Beagle2 supercomputer, Blues and Bebop clusters at Argonne National
Laboratory and the Midway2 cluster at the University of Chicago.
6 SUMMARY AND CONCLUSIONS
Given the expected continuing rapid increases in urban development, the need, as well as the opportunities,
for tools to understand and proactively plan the orderly development of sustainable cities is greater than
ever. Recent and ongoing developments in computing capabilities, both in terms of software and hardware,
the advancement of new modeling paradigms, such as agent-based modeling, combined with the ever-
increasing availability of big data provide unprecedented opportunities for modeling and simulation to make
important and broad impacts on cities. Models can contribute in backcasting (explaining the past),
818

Macal, Collier, Ozik, Tatara, and Murphy
forecasting the future, testing proposed interventions, or explicitly optimizing urban processes. With models
as experimental vehicles and model exploration frameworks facilitating the use of high-performance
computing resources many experiments can be run rapidly and much can be learned quickly. Still, many
challenges remain, as this paper has outlined, from assembling large datasets from disparate sources, to
computational efficiency challenges, to logging models and handling the analytical challenges that large
datasets produced by simulation models generate.
ACKNOWLEDGEMENTS
This work is supported by the U.S. Department of Energy under contract number DE-AC02-06CH11357
and NSF RAPID Award DEB-1516428. This research used computing resources provided on Blues, a high-
performance computing cluster operated by the Laboratory Computing Resource Center at Argonne
National Laboratory, which is a DOE Office of Science User Facility.
REFERENCES
Batty, M., P. Longley, and S. Fotheringham. 1989. “Urban Growth and Form: Scaling, Fractal Geometry,
and Diffusion-Limited Aggregation”. Environment and Planning 21(11):1447–1472.
Batty, M. 2008. “Fifty Years of Urban Modeling: Macro-Statics to Micro-Dynamics.” Edited by Sergio
Albeverio, Denise Andrey, Paolo Giordano, and Dr. Alberto Vancheri. The Dynamics of Complex
Urban Systems: An Interdisciplinary Approach. Physica-Verlag HD.
http://link.springer.com/chapter/10.1007/978-3-7908-1937-3_1.
Bettencourt, L. M. A. 2013. “The Origins of Scaling in Cities”. Science, 340(6139):1438-1441.
Beaumont, M. A. 2010. “Approximate Bayesian Computation in Evolution and Ecology”. Annual Review
of Ecology, Evolution, and Systematics 41(1):379-406.
Collier, N. T., J. Ozik, and C. M. Macal. 2015. “Large-scale ABM with RepastHPC: A Case-study in
Parallelizing a Distributed ABM”. in PADABS 2015 Proceedings. Vienna, Austria, 24-28 August 2015.
Collier, N., and M. North. 2013. “Parallel Agent-based Simulation with Repast for High Performance
Computing”. Simulation 89 (10): 1215–1235.
Dobbs, R./McKinsey Global Institute. 2010. “Megacities”. Foreign Policy Magazine .
Forrester, Jay W. 1969. Urban Dynamics. 1st ed. Pegasus Communication Inc, Cambridge, MA.
Holland, J. H. 1992. Adaptation in Natural and Artificial Systems: An Introductory Analysis with
Applications to Biology, Control, and Artificial Intelligence. 1st ed. Cambridge, Mass: Bradford.
Johnson, R., P. Sydelko, C. Macal, J. Hummel, and M. North. 2016. “Geospatial Analytics Applied to
Urban Environments, Workshop: Mad Scientist 2016: Megacities and Dense Urban Areas in 2025 and
Beyond, April 21-22, 2016, Tempe, Arizona.
Kaligotla C., J. Ozik, N. Collier, C. M. Macal, S. Lindau, E. Abramsohn, and E. Huang. 2018. “Modeling
an Information-Based Community Health Intervention on the South Side of Chicago”. Proceedings of
the 2018 Winter Simulation Conference, edited by M. Rabe et al., Piscataway, New Jersey:IEEE.
Macal, C. M. 2016. “Everything You Need to Know About Agent-based Modeling and Simulation”.
Journal of Simulation 10:144. doi:10.1057/jos.2016.7.
Macal, C. M., M. J. North, N. Collier, V. M. Dukic, D. T. Wegener, M. Z. David, R. S. Daum, P. Schumm,
J. A. Evans, J. R. Wilder, L. G. Miller, S. J. Eells, and D. S. Lauderdale. 2014. “Modeling the
transmission of community-associated methicillin-resistant Staphylococcus aureus: a dynamic agent-
based simulation”. Journal of Translational Medicine 12:124.
Ozik, J., N. T. Collier, J. M.. Wozniak, and C. Spagnuolo. 2016. “From desktop to large-scale model
exploration with Swift/T”. In Proceedings of the 2016 Winter Simulation Conference, 206-220 , edited
by S.Jain et al., Piscataway, New Jersey:IEEE.
Population Reference Bureau. 2013. Food Insight.
819

Macal, Collier, Ozik, Tatara, and Murphy
Settles, B. 2012. “Active Learning”. Synthesis Lectures on Artificial Intelligence and Machine Learning,
6(1):1–114.
Tatara, E., N. Collier, J. Ozik, and C. Macal. 2017. “Endogenous Social Networks from Large-Scale Agent-
Based Models”. IEEE Workshop on Parallel and Distributed Processing for Computational Social
Systems (ParSocial 2017), June 2, 2017, Orlando, FL USA.
Wheaton, W. D., et al. 2009. Synthesized Population Databases: A US Geospatial Database for Agent-
Based Models RTI Press publication No. MR-0010-0905.
Wozniak, J. M., T. G. Armstrong, M. Wilde, D. S. Katz, E.. Lusk, and I. T. Foster. 2013. “Swift/T: Large-
Scale Application Composition via Distributed-Memory Dataflow Processing.” In Cluster, Cloud and
Grid Computing (CCGrid), 2013 13th IEEE/ACM International Symposium, 95–102. IEEE.
AUTHOR BIOGRAPHIES
CHARLES M. MACAL, PhD, PE, is the Director of the Social, Behavioral and Decision Sciences group
at Argonne National Laboratory. He is a member of the INFORMS-Simulation Society, Association for
Computing Machinery, the Society for Computer Simulation International, and the System Dynamics
Society. He has a Ph.D. in Industrial Engineering & Management Sciences from Northwestern and a
Master’s Degree in Industrial Engineering from Purdue. Contact: macal@anl.gov.
NICHOLSON COLLIER PhD, is Software Engineer in the Decision and Infrastructure Sciences Division
at Argonne National Laboratory and Research Staff at the Computation Institute at the University of
Chicago. He is the lead developer of the Repast and RepastHPC(https://repast.github.io) agent-based
modeling toolkits and a core developer of the EMEWS (http://emews.org) framework for large-scale model
exploration. Contact: nick.collier@gmail.com.
JONATHAN OZIK PhD, is a Computational Scientist at Argonne National Laboratory and a Senior
Fellow at the Computation Institute at the University of Chicago. He leads the Repast agent-based modeling
toolkit (repast.github.io) and the EMEWS framework (emews.org) projects. His research focuses on agent-
based modeling and large-scale model exploration. Contact: jozik@anl.gov.
ERIC R. TATARA, PhD, is a Software Engineer at Argonne National Laboratory. He has degrees in
chemical engineering. Contact: tatara@anl.gov.
JOHN T. MURPHY, PhD, is a Computational Social Scientist at Argonne National Laboratory. He has
degrees in anthropology. Contact: jtmurphy@anl.gov.
820