SwarmFest 2014 18th Annual Meeting on Agent-Based Modeling & Simulation University of Notre Dame Notre Dame, IN USA June 29 - July 1, 2014

Conference Center at McKenna Hall Morris Inn

Program Presenters Presentations Abstracts Poster Abstracts Maps

http://www.swarmfest2014.org/ SwarmFest 2014

SwarmFest 2014 Program

Sunday, June 29

5:00-7:00 PM Registration - McKenna Hall

6:00-7:00 PM Women in Computer Science Networking Session - McKenna Hall Room 106

6:00-9:00 PM Reception and Poster Session - McKenna Hall Atrium

9:00 PM Informal Social - Morris Inn, Rohr's Lounge

Monday, June 30 McKenna Hall Auditorium

7:30 AM Continental Breakfast - McKenna Hall Atrium

8:30 AM Welcome - McKenna Hall Auditorium

8:45 AM Keynote: Melanie E. Mose, Ants, T cells and Robots: How does Cooperative Search Emerge in Natural and Engineered Systems?

9:45 AM Break - McKenna Hall Atrium

10:00 AM Mustafa Ilhans Akba and Ivan Garibay: An Initial Agent Based Model for Innovation Ecosystems

Ted Carmichael, Mirsad Hadzikadic, Mary Jean Blink and John C. Stamper: A Multi-Level Complex Adaptive System Approach for Modeling of Schools

11:00 AM Break - McKenna Hall Atrium

11:15 AM Rachel Fraczkowski and Megan Olsen: An Agent-Based Predator-Prey Model with Reinforcement Learning

Russell S. Gonnering and David Logan: Organizational Productivity: Modeling the Interrelationship of Organizational Culture, Intellectual Capital and Innovation

12:15 PM Lunch & Keynote (Morris Inn): Gary An, Evolutionary and Ecological Perspectives on “Systems” Diseases using Agent-based Modeling

1:45 PM Virginia A. Folcik and Gerard J. Nuovo: Finding the Cause of Disease Using Agent-Based Modeling

Erin M. Stuckey: Application of Microsimulation Modeling for Malaria Control Decision-making

2:45 PM Break - McKenna Hal Atrium

3:00 PM Caroline C. Krejci: Structural Emergence in Regional Food Supply Systems

Magda Fontana and Pietro Terna: Agent-based Models Meet Network Analysis: the Policy-making Perspective

Page 1 of 3 SwarmFest 2014

4:00 PM Break - McKenna Hall Atrium

4:15 PM Quirine ten Bosch, Brajendra K. Singh, and Edwin Michael: The Impact of Antibody Dependent Enhancement on Disease Demographics and Transmission Potential of Multi-Serotype Infectious Diseases

Ana Nelson: The Practice of Reproducibility: How Computational Reproducibility Emerges from Researcher Workflow

Megan Olsen and Mohammad Raunak: An Approach to Measure Validation of Agent-Based Simulations

5:45 PM Day's Wrap-up

7:00 PM Dinner - Morris Inn, Fireside Terrance

Tuesday, July 1 McKenna Hall Auditorium

7:30 AM Continental Breakfast - McKenna Hall Atrium

8:30 AM Welcome

8:45 AM Keynote: Michael J. North, Theoretical Analysis of Agent-based Models

9:45 AM Break - McKenna Hall Atrium

10:00 AM S. M. Niaz Arifin, Rumana Reaz Arifin and Gregory R. Madey: Agent-Based Microsimulation (ABμS) Modeling: Revisiting the Micro Perspective

Diggory Hardy and Nakul Chitnis: OpenMalaria, a Simulator of Malaria Transmission and Morbidity, and the Use of BOINC for High-throughput Computing

11:00 AM Break - McKenna Hall Atrium

11:15 AM Scott Christley: Optimal Control of the SugarScape Agent-based Model

Gregory J. Davis and Klaus Kofler: Implementing an Agent-Based Model using OpenCL: A Case Study

12:15 PM Lunch & Keynote (Morris Inn): Greg Madey, Science Gateways: Hosting Agent-Based Models for Use by a Non-modeling Community

1:45 PM Ryan C. Kennedy, Glen E.P. Ropella, and C. Anthony Hunt: A Cell-centered, Agent-based Method that Utilizes a Delaunay and Voronoi Environment in 2- and 3-Dimensions

Will Weston-Dawkes: An Agent Based Modeling Approach to Predicting the Effect Anthropogenic Pressures on the Movement Patterns of Mongolian Gazelles

Page 2 of 3 SwarmFest 2014

2:45 PM Break - McKenna Hall Atrium

3:00 PM Matteo Morini and Simone Pellegrino: Taking Genetic Algorithms and Personal Income Tax Reforms One Step Beyond: Enter Agents

Dave Babbitt and Joel Dietz: Crypto-economic Design: A Proposed Agent-Based Modelling Effort

4:00 PM Break - McKenna Hall Atrium

4:15 PM Md. Zahangir Alam, S. M. Niaz Arifin and M. Sohel Rahman: A Spatial Agent-Based Model of Anopheles vagus for Malaria Epidemiology

R. Ryan McCune & Greg Madey: Emergent Computing with Swarm Intelligent Systems

5:15 PM Day's Wrap-up

6:30 PM Dinner - Morris Inn, Fireside Terrance

Page 3 of 3 SwarmFest 2014 - Presenters

Presenters Title Affiliations

Mustafa Ilhans Akba and Ivan Garibay An Initial Agent Based Model for Innovation Ecosystems University of Central Florida S. M. Niaz Arifin, Rumana Reaz Arifin and Gregory Agent-Based Microsimulation (ABμS) Modeling: Revisiting the Micro University of Notre Dame R. Madey Perspective Md. Zahangir Alam, S. M. Niaz Arifin and M. Sohel A Spatial Agent-Based Model of Anopheles vagus for Malaria Bangladesh University of Engineering & Rahman Epidemiology Technology, Bangladesh and University of Notre Dame Dave Babbitt and Joel Dietz Crypto-economic Design: A Proposed Agent-Based Modelling Effort Northwestern University and University of Pennsylvania Ted Carmichael, Mirsad Hadzikadic, Mary Jean Blink A Multi-Level Complex Adaptive System Approach for Modeling of TutorGen, Inc., University of North Carolina at and John C. Stamper Schools Charlotte, and Carnegie Mellon University Scott Christley, Matthew Oremland, Rene Salinas, Optimal Control of the SugarScape Agent-based Model University of Chicago, Virginia Polytechnic Institute Rachael M. Neilan and Suzanne Lenhart and State University, Appalachian State University, Duquesne University, and University of Tennessee

Gregory J. Davis and Klaus Kofler Implementing an Agent-Based Model using OpenCL: A Case Study University of Notre Dame and University of Innsburck, Austria Virginia A. Folcik and Gerard J. Nuovo Finding the Cause of Disease Using Agent-Based Modeling The Ohio State University and Phylogeny, Inc. Magda Fontana and Pietro Terna Agent-based Models Meet Network Analysis: the Policy-making University of Torino, Italy Perspective Rachel Fraczkowski and Megan Olsen An Agent-Based Predator-Prey Model with Reinforcement Learning Loyola University Maryland Russell S. Gonnering and David Logan Organizational Productivity: Modeling the Interrelationship of Medical College of Wisconsin and University of Organizational Culture, Intellectual Capital and Innovation Southern California Diggory Hardy and Nakul Chitnis OpenMalaria, a Simulator of Malaria Transmission and Morbidity, and Swiss Tropical and Public Health Institute and the Use of BOINC for High-throughput Computing University of Basel, Switzerland Ryan C. Kennedy, Glen E.P. Ropella, and C. Anthony A Cell-centered, Agent-based Method that Utilizes a Delaunay and University of California San Francisco, and Tempus Hunt Voronoi Environment in 2- and 3-Dimensions Dictum, Inc., Portland, OR Caroline C. Krejci Structural Emergence in Regional Food Supply Systems Iowa State University R. Ryan McCune & Greg Madey Emergent Computing with Swarm Intelligent Systems University of Notre Dame Matteo Morini and Simone Pellegrino Taking Genetic Algorithms and Personal Income Tax Reforms One Step Institut Rhônalpin des Systèmes Complexes, ENS Beyond: Enter Agents Lyon, Laboratoire de l’Informatique du Parallélisme, France and University of Torino, Italy Ana Nelson The Practice of Reproducibility: How Computational Reproducibility Dexy and Trinity College, Dublin Emerges from Researcher Workflow Megan Olsen and Mohammad Raunak An Approach to Measure Validation of Agent-Based Simulations Loyola University Maryland Erin M. Stuckey Application of Microsimulation Modeling for Malaria Control Decision- Swiss Tropical and Public Health Institute and making University of Basel, Switzerland

Page 1 of 2 SwarmFest 2014 - Presenters

Quirine ten Bosch, Edwin Michael, and Brajendra K. The Impact of Antibody Dependent Enhancement on Disease University of Notre Dame Singh Demographics and Transmission Potential of Multi-Serotype Infectious Diseases Connor Gibb, Michael Kleyman, Maria Koebel, An Agent Based Modeling Approach to Predicting the Effect University of Maryland Rebecca Natoli, Kyle Orlando, Matthew Rice, Claire Anthropogenic Pressures on the Movement Patterns of Mongolian Weber, Will Weston-Dawkes, Bill Fagan Gazelles

Poster Presenters Title Affiliations

Md. Zahangir Alam, S. M. Niaz Arifin, and M. Sohel A Spatial Agent-Based Model of Anopheles vagus for Malaria Bangladesh University of Engineering & Rahman Epidemiology Technology, Bangladesh and University of Notre Dame Alexander Madey and Holly Goodson Genetic-level Modeling of Directed Yeast Evolution in Turbidostats and University of Notre Dame Chemostats Aboutaleb Amiri, Shant M. Mahserejian, Cameron Computational modeling of bacterial motility and social behavior University of Notre Dame W. Harvey, Morgen E. Anyan, Joshua D. Shrout, and Mark Alber R. Ryan McCune & Greg Madey Decentralized K-Means Clustering: Emergent Computation University of Notre Dame Matthew Staffelbach Lessons Learned from an Experiment in Crowdsourcing Complex Citizen University of Notre Dame Engineering Tasks with Amazon Mechanical Turk

Page 2 of 2 Keynote Speakers

Melanie E. Moses University of New Mexico

Ants, T cells and Robots: How does Cooperative Search Emerge in Natural and Engineered Systems?

Trillions of T cells are flowing through your arteries and crawling through your tissues in their search for pathogens. Without a blueprint of your body or centralized instructions, they protect you from flu, nascent tumors and their own uncontrolled proliferation. Uncountable numbers of ants are crawling across forest canopies, desert sands and maybe your kitchen counter as they search for food. Each species has evolved its own decentralized strategy that tailors a small repertoire of sensing, navigation and communication behaviors to forage effectively in its environment. Spectacularly successful decentralized collective behaviors have evolved in ant colonies and immune systems, but it has proven difficult to engineering effective cooperative systems that can function in the real world. This talk describes what kinds of cooperative search behaviors emerge in ant colonies and immune systems, and how we have replicated some of those behaviors in robotic swarms. We describe what individual behaviors for sensing, navigating and communicating generate robust and efficient cooperative search in different environments, and we discuss implications for natural and engineered complex systems.

BIO: Melanie Moses earned a B.S. from Stanford University in Symbolic Systems, an interdisciplinary program in cognition and computation, and a Ph.D. in Biology from the University of New Mexico in 2005. She is currently an Associate Professor in the Department of Computer Science at the University of New Mexico and External Faculty at the Santa Fe Institute. She continues her interdisciplinary research at the boundaries of Computer Science and Biology with her research lab which includes post docs, and high school, undergraduate and graduate students from Computer Science and Biology. Research in the Moses Lab focuses on computational modeling of complex biological systems, particularly on cooperative search strategies in immune systems and ant colonies. We also apply principles from biology to design computational systems, particularly computer security systems that emulate adaptive immune response and robotic swarms that replicate ant behaviors to perform collective tasks. Dr. Moses co-directs the UNM Program in Interdisciplinary Biological and Biomedical Sciences, co-chaired the Gordon Research Conference on the Metabolic Basis of Ecology, and works with the Santa Fe Institute’s Project GUTS to train teachers and pre-college students in computational modeling of complex systems. She is honored to have been a Ford Foundation Dissertation Diversity Fellow and a Microsoft Research New Faculty Fellowship Finalist, and to have received School of Engineering New Faculty Awards for Teaching and Research.

Michael J. North Argonne National Laboratory

Theoretical Analysis of Agent-based Models

Agent-based modeling has been successfully used to model complex adaptive systems in diverse disciplines. Many of these models were implemented using agent-based modeling software such as Swarm, Repast 3, Repast Simphony, Repast for High-Performance Computing, MASON, NetLogo, or StarLogo. All of these options use modular imperative architectures with factored agents, spaces, a scheduler, and logs. Many custom agent-based models also use this kind of architecture. This talk will introduce and apply a theoretical formalism for analyzing modular imperative agent-based models of complex adaptive systems. The talk will include a discussion of an analytical proof that the asymptotic time and space performance of modular imperative agent-based modeling studies is computationally optimal for a common class of problems. Here ‘optimal’ means that no other technique can solve the same problem computationally using less asymptotic time or space. Given that agent-based modeling is both computationally optimal and a natural structural match for many modeling problems, it follows that it is the best modeling method for such problems. Several other proofs about the time and space performance of modular imperative agent-based models will also be discussed along with validated predictions of the performance of three implemented models.

BIO: Michael J. North, MBA, Ph.D. is the Deputy Director of the Center for Complex Adaptive Agent Systems Simulation within the Decision and Information Sciences Division of Argonne National Laboratory. He is also a Senior Fellow in the joint Computation Institute of the University of Chicago and Argonne. Dr. North has over 20 years of experience developing and applying models for industry, government, and academia. Dr. North has published two books, five conference proceedings, one journal special issue, eight book chapters, 20 journal articles, four invited encyclopedia entries, and over 70 conference papers. Dr. North is also the lead developer of free and open source Repast agent-based modeling suite.

Gary An University of Chicago

Evolutionary and Ecological Perspectives on “Systems” Diseases using Agent- based Modeling

The primary investigative paradigm in biomedicine since the discovery of DNA has been a quest for increasingly detailed characterization of cellular and molecular mechanisms. While this reductionist approach has lead to an explosion in knowledge and data, there is an increasingly poor return in the translation of this mechanistic knowledge into effective clinical interventions. This Translational Dilemma is most pronounced in what can be viewed as “systems” diseases, where the pathophysiology of the disease is due to a failure or dysfunction of a biological control structure. Examples of this type of disorder/disease are cancer, sepsis, hospital-acquired infections, impaired wound healing, auto-immune diseases and diabetes. This talk suggests that the essential qualities of these disease processes are insufficiently captured with the current protein-gene-pathway-centric focus of the biomedical research community (including the fields of systems and computational biology), in large part because these approaches do not generally account for biology’s primary theory, evolution, or the description of biology’s primary context, ecology. I suggest that the effective modulation of “systems” diseases will need to involve their characterization in terms of evolutionary and ecological dynamics. Given its heritage in the fields of Artificial Life and Ecology, Agent-based modeling will play a critical role in providing the needed evolutionary and ecological context for these pathophysiological processes. This talk will present specific examples of agent-based modeling used to examine the evolutionary dynamics of cancer and the ecology of Clostridum difficile colitis.

BIO: Dr. Gary An is an Associate Professor of Surgery and the co-Director of the Surgical Intensive Care Unit at the University of Chicago. In addition to being an active clinician he is a Senior Fellow of the Computation Institute at the University of Chicago. He is a graduate of the University of Miami, Florida School of Medicine, and did his surgical residency at Cook County Hospital/University of Illinois, Chicago. He is a founding member of the Society of Complexity in Acute Illness (SCAI) and is the current president of the Swarm Development Group, one of the original organizations promoting the use of agent-based modeling for scientific investigation. He is the founder and director of the Fellowship in Translational Systems Biology at the University of Chicago. He is member of multiple medical and computer science societies, and serves on the editorial board of several journals. He has worked on the application of complex systems analysis to sepsis and inflammation since 1999, primarily using agent based modeling to create mechanistic models of various aspects of the acute inflammatory response, work that has evolved to the use of agent-based models as a means of dynamic knowledge representation to integrate multiple scales of biological phenomenon. The impetus for his work is the recognition that the Translational Dilemma has arisen from a bottleneck in the scientific cycle at the point of experiment and hypothesis evaluation. His research involves the development of: mechanism- based computer simulations in conjunction with biomedical research labs, high- performance/parallel computing architectures for agent-based models, artificial intelligence systems for modular model construction, and community-wide meta-science environments, all with the goal of facilitating transformative scientific research.

Greg Madey University of Notre Dame

Science Gateways: Hosting Agent-Based Models for Use by a Non- modeling Community

Science Gateways are collections of online services that enable select communities of users to access tools, applications, data collections, workflows, visualizations, resource discovery and compute resources. One such Science Gateway under development is VecNet, the Vector-Borne Disease Network. VecNet, sponsored by the Bill & Melinda Gates Foundation, will initially serve a community of researchers, product developers, public- health managers, funding agencies, and policy makers focused on the goal of global eradication of Malaria. The VecNet Science Gateway is unique in that its primary function is the hosting of two complex agent-based models for use by non-modelers. While other models may be added in the future, the two models being deployed now are OpenMalaria and EMOD. The VecNet architecture, the two agent-based models, design challenges, and its current status will be described.

BIO: Dr. Greg Madey is a Professor in the Department of Computer Science & Engineering, College of Engineering, University of Notre Dame. His formal education includes degrees in Mathematics and Operations Research. He has worked with microsimulation, discrete- event simulation, individual-based modeling and agent-based modeling for over 30 years. He was an early user of the Swarm agent-based modeling toolkit (late 1990's) and hosted two prior SwarmFest Meetings at the University of Notre Dame in 2003 and 2006. His recent research projects using agent-based modeling include: a study of the social-networks of open source software developers and projects, the behavior of natural organic matter in the soil and water, investigations into emergency management using dynamic data driven agent-based simulations, disease transmission among macaque monkeys in Bali, investigations into the design of command and control for UAV swarms, and most recently, modeling of malaria transmission. He has supervised approximately 30 Ph.D. dissertations to completion, with several more under way.

SwarmFest 2014

Extended Abstracts

An Approach to Measure Validation of Agent-Based Simulations

Megan Olsen Mohammad Raunak Department of Computer Science Department of Computer Science Loyola University Maryland Loyola University Maryland Baltimore, MD 21045 Baltimore, MD 21045 [email protected] [email protected]

Modeling and simulation is a primary approach for studying complex systems, where one creates an abstraction of the real-world system to be studied. It can be a powerful tool for safely and efficiently discerning the likely outcome of a decision. For many years the primary simulation approach was discrete event, abstracting problems into a series of timed events. Although this approach is still effective in many domains, with increased computational and algorithmic capabilities more modeling options are now available. Agent-based modeling has become a clear favorite in the research community in recent years, as evident from the increase in the number of papers using this modeling technique. ‘Validation’ is the primary mechanism for assessing the correctness and usefulness of a simulation model. To date, there is no standard for quantifying the level of validation of a simulation, although reviewers and audience members are often quick to ask, “how did you validate the model?” We currently have no concrete approach for discerning whether the validation performed is sufficient, which can be applied systematically to all agent-based simulations. Likewise, we lack a specific approach for applying a set of validation techniques systematically over all agent-based models. Although the use of simulation is continuing to grow, some fields such as cancer biology currently struggle to accept that simulations can provide meaningful representation of their field’s problems (Kitano, 2002). A concrete measure of validation is necessary to continue increasing the applicability and trust of simulation, especially for fields such as biological systems. We propose a quantification of performed validation on simulations called a “validation coverage criterion.” The intent of the validation coverage criterion is to measure the extent to which a simulation has been evaluated against its intended behavior. To calculate validation coverage we propose the following parallel processes to the current validation approach: 1. Define the validatable elements of the executable simulation model. 2. Determine applicable techniques for validating each element. 3. Track what validation techniques have been successfully applied. 4. Calculate the coverage obtained through step 3. To accomplish these steps, a number of questions must be answered: (a) How do we ensure that all elements to be validated have been determined? (b) How do we determine the techniques for validation for each element? (c) Can we ensure objectivity in user tracking of validation application and its success? and, (d) How do we compute the coverage? In our prior work we proposed an initial approach for the validation coverage process for agent-based simulations to answer questions (a) and (d) above (Olsen, 2013). In our approach, we guide the simulation modelers to discover all validatable elements through the systematic consideration of different ‘aspects’ and ‘element types’ in any agent-based simulation model. We compute the validation coverage based on how well these identified elements in each of the aspects have been validated, depending primarily on how many different validation techniques were applied. However, this work had some shortcomings that we now address. We first improve the calculation of validation coverage such that the importance of each validatable element of an agent-based simulation (such as a reproduction rate, or movement probability) can be specified, instead of focusing on the relative importance only at the category level, i.e. the “aspects.” Second, we propose a technique for determining the importance of applicable validation techniques to increase the objectivity of the coverage calculation. The ‘validation coverage’ metric and the process of computing it helps us quantify and assess how much validation has been done on a simulation. It also provides guidelines that may be used by both practitioners and users/customers across many types of simulation systems. The validation coverage criterion could affect many scientific fields, as simulation has become a common tool in engineering, psychology, ecology, cellular biology, sociology, and more. This criterion could provide agent-based modelers, including the Swarmfest community, the ability to ensure proper validation, as well as increase the adoption of simulation in fields that have previously been concerned that simulation results lack adequate validation proof.

References Kitano, H. (2002). Systems Biology: A brief overview. Science, 295, 1662-1664. Olsen, & Raunak. (2013). A Framework for Simulation Validation Coverage. Proceedings of the Winter Simulation Conference.

The Practice of Reproducibility: How computational reproducibility emerges from researcher workflow.

Ana Nelson

Extended Abstract:

This will be a practical talk showing a reproducible process for generating a mock research paper based on an agent-based simulation using open source tools like Docker (docker.io) and Dexy (dexy.it).

Agent-Based Modelling researchers have more to gain than most programmers by employing good reproducible development and documentation practices. Not only do ABM researchers have to maintain simulation code, they have to analyze generated data and write up results, with the possibility of having to completely redo this analysis if they update their models and generate new data. A reproducible workflow based on open source tools not only benefits the individual researcher or team using it, it also benefits the whole community, since anyone else can easily reproduce results and re-use code.

This talk will describe one example of a fully automated (and therefore reproducible) workflow, including downloading and installing all necessary software and simulation source code, running the simulation to generate data, running analysis scripts to generate plots and calculated data, and embedding these into documents. We will see how changing simulation code and re-running the workflow results in automatically updated documents: plots will change to reflect the newly-generated data, and software documentation will show the updated source code.

There are two goals for this talk. First, to make researchers aware of the possibilities of a fully automated and reproducible workflow. Second, to present one possible suite of open source tools for implementing such a workflow.

Dexy (dexy.it) is a tool for automating projects. It's similar in spirit to GNU make, but with lots of document-related features. Dexy can automate running scripts and processes, and it also facilitates the embedding within documents of generated artifacts like source code, data and plots. It's easy to add Dexy to an existing project, and Dexy supports multiple document formats and programming languages. It's a software documentation tool and a reproducible research tool, and it was directly inspired by the challenges of Agent-Based Modelling research.

Docker (docker.io) is a convenient tool for creating isolated lightweight "containers" on Linux. By creating a separate container for each project you can keep your primary operating system clean of clutter and also create a reproducible workspace for each project. Dockerfiles, used to configure the setup script for each container, are easily readable by non-Docker users, so they can act as either an executable script, or as a testable list of software dependencies.

Speaker Info:

Ana Nelson works as a software consultant in the San Francisco Bay Area. Previously, she completed a Ph.D. in economics at Trinity College, Dublin and attended Swarmfest as a graduate student. She is the author of the open source Dexy package for project and document automation.

SwarmFest 2014, University of Notre Dame

Matteo Morini* Simone Pellegrino°

Taking Genetic Algorithms and Personal Income Tax Reforms One Step Beyond: Enter Agents.

Abstract:

The authors' initial research on fiscal systems optimization implied a static situation where taxpayers did not react to adjustments to the tax structure impacting their personal income. In a real­world prime example, a given reduction of the tax revenue (decided by the Italian government in 2014) was to be spread in form of tax savings among taxpayers. The equi­ table goal to be attained was the maximization of the tax redistributive effect, while pre­ venting all taxpayers to be worse off with respect to the present tax structure. To this end, we employed a genetic algorithm to search the huge combinatorial space resulting from >30 parameters (marginal tax rates, income thresholds, allowances and deductions, tax credits...). The original model implemented applies to the short term, but it may be the case (and a vast literature is there to demonstrate) that taxable income responds to tax rates in the medium­long term. We try and incorporate behavioural response by the taxpayers as agents' elasticities, in a dynamic model where each tax structure adjustment is followed by a heterogeneous shift in the individual earning decisions, ...and back to square one.

* IXXI, Institut Rhônalpin des Systèmes Complexes, ENS Lyon, Laboratoire de l’Informatique du Parallélisme, INRIA­UMR 5668, France and Department of Economics and Statistics, University of Torino, Italy

° Department of Economics and Statistics, University of Torino, Italy EMERGENT COMPUTING WITH SWARM INTELLIGENT SYSTEMS

R. Ryan McCune & Greg Madey University of Notre Dame Computer Science & Engineering Department Notre Dame, IN 46556

Abstract

Challenges posed by Big Data exemplify the limits of centralized systems, limits that arise from two properties. First, inherent to centralized systems are bottlenecks, which limit scalability while exposing the system to cascading failure. Second, centralized systems often require global information, leading to intractable computation. These limits, inherent to centralized design, necessitate a new paradigm for future computational systems.

Emergent computation may potentially alleviate the challenges of Big Data, and lies at the intersection of distributed computing and swarm intelligence. Distributed computing describes a system of interconnected computers cooperating to accomplish a task. Swarm intelligence characterizes a multi-agent system that utilizes emergent behavior to solve problems. Swarms are decentralized, and self- organize into a structure that is robust, scalable, adaptable, and computationally efficient. In swarm intelligent systems, or swarms, simple and local behaviors distributed across many agents lead to a complex global phenomena. Emergent computing occurs when the emergent behavior is also a computation.

While emergent computation has yet to be fully realized, two swarms are presented that perform a computation outside of a distributed computing environment. The first swarm is based off a well-known ant foraging model where ants collectively uncover the shortest path to food. The ant foraging model is adapted to a second model, where agent behaviors result in an emergent clustering. The swarms are evaluated with agent-based models. An improved understanding of emergent computation is realized by exploring swarms that perform a computation. The swarms are later adapted to a distributed computing environment.

Structural Emergence in Regional Food Supply Systems Caroline C. Krejci Department of Industrial and Manufacturing Systems Engineering Iowa State University

The modern industrial food supply chain (FSC) is extraordinarily productive but faces serious long-term environmental and social sustainability challenges. Unsustainable resource consumption (e.g., fossil fuels and water) and ecological degradation (e.g., agrochemical runoff, greenhouse gas emissions, soil erosion, wildlife habitat destruction, and biodiversity loss) are problems that have been further compounded by climate-change-induced precipitation and temperature variability, as well as changes in the frequency and severity of extreme climate events. Such variability is particularly troubling for the industrial food system, which is characterized by significant consolidation and centralization at all echelons (e.g., producers, processors, distributors, and retailers). While in some respects this type of network structure can be considered highly efficient, it is also inflexible and vulnerable to supply disruptions – when capacity is located in just a few centralized locations, the lack of overall structural complexity and redundancies increases the risk of system failure due to single component failure (e.g., drought in a food-producing region). This streamlining of the FSC has also destroyed the livelihoods of many small- and medium-scale farmers, as well as the rural communities that they inhabited, and it has significantly increased transport distances between producers and consumers (i.e., “food miles”). Maintaining the food system’s long-term productivity without compromising environmental and public health and well-being, particularly in the face of increasing external variability, is a major challenge. Producers and consumers have responded to these growing concerns in a variety of ways. In the U.S., there has been a rapid growth of the local/regional food movement, in which consumers prefer food that is produced in geographic proximity to them (e.g., through farmers’ markets), rather than from distant sources. Consumers value the perceived quality and safety of local products, the availability of information about the producers and their production methods, and the feeling of supporting regional business. By selling directly to local consumers (rather than to mainline distributors), producers often benefit from higher prices and lower volume requirements. However, direct sales through farmers’ markets are inefficient for larger-scale producers and institutional customers. To address this problem, there have been recent efforts to develop farm-to-institution marketing/distribution channels, such as regional food hubs. Food hubs act as regional aggregation points between producers and institutional buyers, for both physical products and information (e.g., inventory and order status). In this way, institutional buyers gain access to regionally-produced food without overwhelming transaction costs, and medium-scale producers gain access to a market with consistent demand and adequate prices. However, the process of food hub development and FSC regionalization has faced many challenges. In particular, it requires significant vertical and horizontal coordination among FSC actors. The future of FSC regionalization and its impact on long-term FSC sustainability and resilience is unclear. The following research questions are of interest:

 What are the impacts of different policies (e.g., incentives, regulations) on the emergence of different types of FSC structures (e.g., farmer collectives, coordinated buyer-supplier relationships, food hubs) over time?  How do these emergent structures impact long-term FSC sustainability outcomes and resilience to external variability?  What are the implications of these outcomes for FSC actors (e.g., economic viability, food security)? Improving farm-scale environmental sustainability has received significant attention in the modeling literature. However, very little existing work has examined overall FSC behavior, particularly with respect to social and economic sustainability over time. Food systems (and supply chains in general) are social systems, with system-wide outcomes that depend upon the behavior of and interactions among constituent actors. Traditional analytical and discrete-event simulation modeling techniques are unable to capture the dynamic and complex interactions, adaptations, and behaviors of individual FSC actors, which is essential to gaining a better understanding of FSC structural development. Agent-based modeling (ABM) is a tool that is particularly well-suited to modeling such systems. Therefore, to address the aforementioned research questions, we have developed an ABM of a theoretical FSC in NetLogo. Our model contains four agent types: farmers, farmer collectives, distributors, and institutional customers. Each agent exists in one of four distinct geographical regions, where regionalization is defined as trade that occurs entirely within a given region. In each time-step (season), farmers determine whether they want to work independently or join/form a collective, based on their personal attributes/preferences and current environmental conditions. Each farmer agent then selects and produces a crop (subject to weather and regional variability), harvests it, and negotiates with potential buyers in an attempt to sell it for the largest possible profit. Farmers can sell to distributors in any region, and these distributors then sell crops to institutional customers within their own regions. Farmers can also sell directly to institutional customers within their own regions, although this often requires significantly greater transaction and transportation costs. We have used this model to examine the relationships among experimental parameter values, FSC structural development, and FSC outcomes. In particular, our research has focused on 1) horizontal coordination decisions among farmer agents, who can collectively aggregate yields of a single crop type in order to achieve greater volumes and receive higher prices from buyers, and 2) vertical coordination between farmer agents and customer agents, in which supplier selection decisions are based on customers’ preferences for regionally-produced food, convenience, and low cost. Our current work is focused on the development of increasingly complex FSC agent coordination mechanisms to enable the emergence of new FSC structures (e.g., food hubs): farmers and buyers will be able to develop contracts and longer-term relationships, farmers will be able to collectively coordinate crop planning activities to better meet customer demand, and distributors will be able to respond strategically to customers’ preferences and behaviors (to avoid being cut out of the FSC via direct farmer-customer transactions). We are interested in using this theoretical model to gain a better understanding of the types of coordination and network structures that develop under certain conditions (e.g., policy implementations, environmental conditions), and the implications of these structures for system sustainability outcomes.

2

A Cell-centered, Agent-based Method that utilizes a Delaunay and Voronoi Environment in 2- and 3-Dimensions

Ryan C. Kennedy1, Glen E.P. Ropella2, and C. Anthony Hunt1

such as the extent of a cell-entity. D/V grids give us Short Abstract — We present an agent-based method that additional capabilities to generate behaviors from the agent’s characterizes cell-behavior from a cell’s perspective in an off- perspective, utilizing neighborhood-based information. We lattice environment. A goal is to facilitate study of more represent complex entities and behaviors at an agent-level. advanced cell behavior through a dynamic Delaunay and In ABM, each entity, or agent, implements a set of rules and Voronoi off-lattice environment, in both 2- and 3-dimensions. We demonstrate the method using cell-entities that map to has its own set of properties. Numerous agent classes can living cells. Entities are also agents. Use cases are highlighted exist within a model, and the interactions between agents and we expand on existing cell- and agent-centered methods by can lead to emergent properties. Because mechanisms in offering a new perspective. As the demand for biomimetic ABM are often informally specified, with few or no models grows, new methods, such as the one described, will be theoretical constraints, natural phenomena can be simulated needed to improve mechanistic explanations of biomedical more directly, in contrast to those methods that require phenomena. arching principles which may bias simulation results. Keywords — biomimetic, off-lattice. Further, ABM lends itself to modeling cell behavior. We are applying our method to biological cells and have simulated I. EXTENDED ABSTRACT basic coarse grain cell behavior from a cell’s perspective, utilizing 2- and 3-dimension D/V grids ew simulation methods are enabling models to become We demonstrate the method by expanding upon more biomimetic. Cell-centered models are among the N traditional cell- and agent-centered methods. So doing opens most popular. There is a pressing need for enhanced the door to several use cases. Representing space with D/V methods to improve explanatory mechanistic insight into grids has potential as an alternative to lattice-based biomedical health-morbidity phenomena. Agent-based environments and is advantageous in a number of modeling (ABM) is adept at simulating natural phenomena. applications, such as in modeling wireless networks. Our We report progress toward a cell-centered, agent-based method is intended for application across domains, as our method that utilizes an agent’s perspective, looking out at cell-entities can represent or map to any referent entity, at a the environment. Representing space as Delaunay and variety of granularities. We foresee this method providing Voronoi (D/V) grids facilitates achieving that perspective. additional understanding, in particular, when utilized to Simulating from an agent’s perspective allows us to observe improve biomedical insight into various health related and validate from both the agent- and system-levels. phenomena. Among cell-centered methods, the cellular Potts model (CPM) is the most widely used and extensible technique to REFERENCES simulate a variety of cell behaviors, such as morphogenesis, adhesion, and virtual tumors [1-5]. In the CPM, cells exist [1] Merks RMH, Glazier JA (2005) A cell-centered approach to developmental biology. J Phys A 352(1), 113-130. within a lattice and are connected with bonds represented by [2] Savill NJ, Hogeweg P (1997) Modelling morphogenesis: From single various equations; cell behaviors are governed largely by cells to crawling slugs. J Theor Biol 184(3), 229-235. such energies. Our method differs from the CPM in its [3] Hogeweg P (2000) Evolving mechanisms of morphogenesis on the interplay between differential adhesion and cell differentiation. J perspective. The CPM relies on a grid-based perspective at a Theor Biol 203, 317-333. system-level. By allowing an agent to observe its [4] Zhang L, Athale CA, Deisboeck TS (2007) Development of a three- neighborhood, we allow cell-entity agents to utilize local dimensional multiscale agent-based tumor model: Simulating gene- protein interaction profiles, cell phenotypes and multicellular patterns information to inform behavior in a dynamic, off-lattice in brain cancer. J Theor Biol 244(1), 96-107. environment. In an off-lattice spatial representation, [5] Voss-Böhme A (2012) Multi-scale modeling in morphogenesis: a continuous values are used to represent spatial boundaries, critical analysis of the cellular Potts model. PLoS One 7(9), 1-14.

1Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco, CA, USA. E-mail: [email protected], [email protected] 2 Tempus Dictum, Inc., Portland, OR, USA. OpenMalaria, a simulator of malaria transmission and morbidity, and the use of BOINC for high-throughput computing

Hardy, D., Chitnis, N., and colleagues Associations: Swiss TPH ,∗ University of Basel† Contact: [email protected]

Abstract Malaria is an infectious disease, spread through mosquito bites, and responsible for substantial morbidity and mortality, principally in sub- Saharan Africa. In the last decade significant reductions in transmission and burden have been achieved; these gains, however, now face the twin threats of decreased funding for control and the development of resistance to both drugs and mosquito-targeting interventions. Mathematical models can be useful in planning the deployment of cur- rent interventions and developing new tools. We present an agent-based model of malaria in humans, OpenMalaria [1], which includes demogra- phy, heterogeneity, dynamics of individual infections, and acquired immu- nity [2,3], and is linked to a population-based model of seasonal malaria transmission in mosquitoes [4]. The OpenMalaria code is open source under the GNU Public License, and has been developed since 2004 in collaboration between the Swiss TPH and the Liverpool School of Tropical Medicine. It relies on several free software tools, among them Berkeley’s volunteer computing platform (BOINC), the C++ Boost libraries and the GNU Scientific Library. OpenMalaria allows the simulation of the deployment of multiple con- trol interventions concurrently with independent decay rates and func- tions, and outputs both transmission levels and clinical event statistics. Examples of intervention scenarios include combinations such as insec- ticidal treated nets along with improvements in access to official health care [5], or the introduction of vaccines in a setting with existing cov- erage of indoor residual spraying [6]. Outputs include predictions of ef- fectiveness in reducing number of clinical cases (sicknesses and deaths), disability-adjusted life years, prevalence (detectible infections), and inoc- ulation rates. The talk will first focus on an overview of our malaria model before moving on to discuss how we use the BOINC computing platform to cope with large experiments (involving tens of thousands to millions of individ- ual simulations, each involving tens to hundreds of thousands of agents) and parameter uncertainty (genetic algorithms to improve a statistical measure of fitness).

∗Swiss Tropical and Public Health Institute, Socinstr. 57, 4051 Basel, Switzerland †University of Basel, Petersplatz 1, 4003 Basel, Switzerland

1 References

[1]“ OpenMalaria,” http://code.google.com/p/openmalaria/, date accessed: 2 June 2014. [2] T. Smith, N. Maire, A. Ross, M. Penny, N. Chitnis, A. Schapira, A. Studer, B. Genton, C. Lengeler, F. Tediosi, D. de Savigny, and M. Tanner, “Towards a comprehensive simulation model of malaria epidemiology and control,” Parasitology, vol. 135, pp. 1507–1516, 2008. [3] T. Smith, A. Ross, N. Maire, N. Chitnis, A. Studer, D. Hardy, A. Brooks, M. Penny, and M. Tanner, “Ensemble modeling of the likely public health impact of the RTS,S malaria vaccine,” PLoS Medicine, vol. 9, no. 1, p. e1001157, 2012. [4] N. Chitnis, D. Hardy, and T. Smith, “A periodically-forced mathematical model for the seasonal dynamics of malaria in mosquitoes,” Bulletin of Math- ematical Biology, vol. 74, no. 5, pp. 1098–1124, 2012.

[5] O. J. T. Bri¨etand M. A. Penny, “Repeated mass distributions and contin- uous distribution of long-lasting insecticidal nets: modelling sustainability of health benefits from mosquito nets, depending on case management,” Malaria Journal, vol. 12, p. 401, Nov. 2013.

[6] T. A. Smith, N. Chitnis, O. J. T. Bri¨et, and M. Tanner, “Uses of mosquito-stage transmission-blocking vaccines against Plasmodium falci- parum,” Trends in Parasitology, vol. 27, no. 5, pp. 190–196, May 2011.

2 Organizational Productivity: Modeling the Interrelationship of Organizational Culture, Intellectual Capital and Innovation

Russell S. Gonnering* David Logan**

*Department of Ophthalmology, The Medical College of Wisconsin, 1780 San Fernando Drive, Elm Grove, WI 53122 [email protected] **Marshall School of Business, The University of Southern California, 3670 Trousdale Parkway, Los Angeles, CA 90089 [email protected]

Abstract. Improvement of organizational performance is a near universal, yet tantalizingly elusive, goal. We have developed a NetLogo agent-based model that is significantly different from prior models of culture. It explores the nonlinear modulation of organizational productivity through the interrelationship between organizational culture, intellectual capital, shared values and common purpose. The model builds upon a prior presentation in which a similar model confirmed that culture spreads through an organization in meme-like fashion and that cultural propagation is highly dependent on upon the initial state of the culture. This new model mirrors the known phase-transition between stages of culture, the critical impact of shared and resonant core values on performance and the striking non-linear jumps in productivity as the culture shifts. Special emphasis is placed on the influence of: 1) formation of triads (closing “Structural Holes” in the organization) as a prime tool to effect cultural advancement and increase in organizational productivity; 2) effects of coalescing values and purpose; and 3) innovation. This approach to performance improvement differs from the more common focus upon the “hard” aspects of organizations—processes, strategy and structure—which have produced disappointing gains. The model demonstrates allometric scaling with appropriate utilization constants as a means of understanding nonlinear jumps in organizational productivity, with intellectual capital in the organization analogous to mass in the organism. Graphing attractors shows a striking difference between operation of the model at baseline and when structural holes are closed:

Keywords: agent-based modeling; organizational culture; productivity; Tribal Leadership; performance improvement; Complex Adaptive System; Structural Hole; intellectual capital; allometric scaling An Agent-Based Predator-Prey Model with Reinforcement Learning

Rachel Fraczkowski Megan Olsen Department of Computer Science Department of Computer Science Loyola University Maryland Loyola University Maryland Baltimore, MD 21045 Baltimore, MD 21045 [email protected] [email protected]

In population dynamics we generally analyze either how a single population changes over time, or how two populations interact and influence their population sizes over time. Population dynamics are studied using a wide variety of computational models, from differential equations such as Lotka-Volterra, to individual-based models such as either cellular automata or agent- based models (ABM). Agent-based simulations allow the agents to learn from their experiences, and adapt their behaviors so they are better suited to their environment. With that level of flexibility, agent- based modeling allows for a variety of discipline applications. From modeling insect behavior (Luke, 2005) to modeling the fall of ancient civilizations (Kohler et al, 2005), agent-based modeling brings understanding and clarity to complex natural interactions by allowing them to develop through the individuals. This individual development can provide added realism over other modeling approaches. We explore the evolution of agents in a predator-prey system. Although ABM has been used in population dynamics previously, the agents rarely learn from their experiences. We study population dynamics with evolution, where agents in both the predator and prey populations learn from their experiences in the environment. Our agents learn through TD-learning, a type of reinforcement learning. Reinforcement Learning (RL) is a machine learning technique based on the psychological area of study of the same name. It is very similar in premise: the desired behaviors are rewarded, whereas the undesirable behaviors are either ignored or punished. An RL-agent has a goal state that it is trying to reach. After attempting some action, the agent will receive a value or level of reward or punishment (negative reward). The agent tries to maximize the sum of these reinforcement values from the initial state until its end state. The mapping from state or action to value allows for reinforcement of certain actions over others, until the agent learns the desired behavior. We analyze our agent-based learning model of predator and prey in three scenarios: only the prey learn, only the predators learn, and both species learn. In addition to prey and predator we also have a non-evolving food source. The food source can be removed from the grid by prey and replenishes slowly over time. Agents move by biased random movement in their Moore neighborhood, and update their probability of movement based on the reinforcement from their previous action. Agents pass their learned biases to the new generation. Our results show that in all three cases each species is able to evolve to show increases in the expected behavior. We observe the expected chasing patterns of predator-prey in visualization, and movement learning on the agent level: prey learn to avoid predators, and predators learn to chase prey through rewards. With the introduction of food growing in clusters instead of randomly, prey learn to stay in the food areas unless threatened by a predator. We also observed the highest learning rate (70.8%) occurring when a species’ population was at its highest level, and the lowest learning rate (<1%) when a species’ population was at its lowest level. The results of our work indicate that reinforcement learning can be beneficial in population dynamics models to increase the realism of the model. We show through our framework how to successfully use TD-learning for agent-based simulations in which agents must learn how to move in their world without moving toward a specific goal location. We are unaware of any other research resulting in an agent-based population dynamics model using TD- learning in which the agents are learning general movement strategies in response to actions taken by competitor agents. We propose that this framework can be incorporated into other agent-based models in which learned movement habits are desired.

References

Luke, S., Cioffi-Revilla, C., Panait, L., Sullivan, K., Balan, G. (2005). MASON: A Multi-Agent Simulation Environment. Simulation: Transactions of the society for Modeling and Simulation International. 82(7), 517-527.

Kohler TA, Gumerman GJ and Reynolds RG (2005). Simulating ancient societies. Scientific American. 293(1): 77–84.

Magda Fontana and Pietro Terna

University of Torino, Italy

Agent-based models meet network analysis: the policy-making perspective

An important perspective use of Agent-based models (ABMs) is that of being employed as tools to support decision systems in policy-making, in the complex systems framework. Such models can be usefully employed at two different levels: to help in deciding (policy- maker level) and to empower the capabilities of people in evaluating the effectiveness of policies (citizen level). As a consequence, the class of ABMs for policymaking needs to be both quite simple in its structure and highly sophisticated in its outcomes.

The pursuing of simplicity and sophistication can be made more efficacious by applying network analysis to the emergent results. As a matter of fact, in the actual world the consequences of choices and decisions and their effects on society, and on its organization, are equally relevant.

Considering together the agent-base and network techniques, we have a further important possibility. Being easier to have network data (i.e. social network data) than detailed behavioral individual information, we can try to understand the links between the dynamic changes of the networks emerging from agent-based models and the behavior of the agents. As we understand these links, we can apply them to actual networks, to guess about the content of the behavioral black boxes of real-world agents.

We propose a simple basic structure where events, scheduled upon time, ask agents to behave, to modify their context, and to create new structures of links among them. Events are organized as collections of small acts and steps. The metaphor is that of a recipe, i.e. a set of directions with a list of ingredients for making or preparing something, especially food (as defined in the American Heritage dictionary). Technically, recipes are sequences of numerical or alphanumerical codes, reported in vectors, and move from an agent to another determining the events and generating the edges of the emerging networks.

A basic code will be shown and several examples in different fields will be suggested: production, health-care scenarios, paper co-authorship, opinion spreading, etc. Finding the cause of disease using Agent-Based Modeling

Virginia A. Folcik, Ph.D.1,2,3 and Gerard J. Nuovo, M.D.4

The Ohio State University (OSU) at Marion1, OSU Computer Science and Engineering2, The OSU Innovation Group for the Study of Complexity in Human, Natural, and Engineered Systems3, and Phylogeny, Inc.4

Background: An agent-based model called the Basic Immune Simulator 2010 (BIS_2010; [1]) was modified to study the role of the immune system in idiopathic pulmonary fibrosis (IPF). IPF is a lethal, restrictive lung disease with a life expectancy of two to three years post-diagnosis, regardless of treatment [2]. The generic virtual tissue space of the BIS_2010 was converted to lung tissue. This process involved extensive literature searches to compile information that could be programmed as lung cell-agent behavior in the model. Additional laboratory experiments were needed to determine the immune cell types present in the lung, including recently characterized T-lymphocytes called T-helper-17s [3, 4], and to characterize the cytokines present (signals produced by cells of the immune system). These experiments yielded unexpected results [5]. Some cytokines (IL-17) and other biological molecules were found marking cells that are unique to the pathology of pulmonary fibrosis, instead of in the T-helper-17s where one would expect to find them. These findings raised questions about the origin of the disease-specific lung cells, and even the IL-17 itself. IL-17 was originally discovered in activated T-lymphocytes over two decades ago [6], but its significance in inflammation was not realized until a decade ago [7]. At its discovery, its homology to a gene in Herpesvirus saimiri was noted [6]. It was soon after confirmed to be a mammalian cytokine stolen by Herpesvirus saimiri [8], most closely homologous to murine or rat IL-17 [6]. Herpesvirus saimiri contains the most pirated mammalian genes of any DNA-virus sequenced to date [9].

Results: Given the unexpected results from the immunological survey of the IPF lung tissue, the question that begged to be answered was whether Herpesvirus saimiri could be found in samples of lung tissue from IPF patients. This would explain many characteristics of IPF, including its age of onset (similar to the age of varicella-zoster reactivation), and temporally heterogeneous nature (another feature attributable to sporadic Herpesvirus reactivation). Herpesvirus saimiri was present in the lung tissue from IPF patients, but not in lung tissue from patients with other fibrotic lung diseases. It was clearly present in the unusual epithelial cells unique to the pathology of pulmonary fibrosis, where it colocalized with IL-17 and three other mammalian proteins known to be in the Herpesvirus saimiri genome [10]. This discovery made further development of the agent-based model, the BIS- Lung, unnecessary. The model had served its purpose for this project. It was also unfunded for more than a year at the time of the discovery. Any further development will be spurred by a new purpose.

Conclusions: Several points known to agent-based modelers have been demonstrated by this study [11]:

a) When the information included in an agent-based model is chosen carefully and is at the appropriate level of detail, the model can lead the investigator to the answer to the problem, albeit indirectly. The model can define the questions requiring laboratory investigation.

b) An agent-based model need not be completed to be fruitful.

c) Agent-based modeling helps to solve problems regarding complex systems, including what causes disease.

References: 1. Folcik, V.A., et al., Using an agent-based model to analyze the dynamic communication network of the immune response. Theoretical Biology and Medical Modelling, 2011. 8: p. 1. 2. Rafii, R., et al., A review of current and novel therapies for idiopathic pulmonary fibrosis. Journal of Thoracic Disease 2013. 5(1): p. 48 - 73. 3. Harrington, L.E., et al., Interleukin 17-producing CD4+ effector T cells develop via a lineage distinct from the T helper type 1 and 2 lineages. Nature Immunology, 2005. 6: p. 1123-1132. 4. Park, H., et al., A distinct lineage of CD4 T cells regulates tissue inflammation by producing interleukin 17. Nature Immunology, 2005. 6: p. 1133-1141. 5. Nuovo, G.J., et al., The distribution of immunomodulatory cells in the lungs of patients with idiopathic pulmonary fibrosis. Modern Pathology, 2012. 25: p. 416-433. 6. Rouvier, E., et al., CTLA-8, cloned from an activated T cell, bearing AU-rich messenger RNA instability sequences, and homologous to a Herpesvirus Saimiri gene. The Journal of Immunology, 1993. 150(12): p. 5445-5456. 7. Weaver, C.T., et al., IL-17 family cytokines and the expanding diversity of effector T cell lineages. Annual Review of Immunology, 2007. 25: p. 821-852. 8. Yao, Z., et al., Herpesvirus Saimiri encodes a new cytokine, IL-17, which binds to a novel cytokine receptor. Immunity, 1995. 3: p. 811-821. 9. Albrecht, J.-C., et al., Primary structure of the herpesvirus saimiri genome. Journal of Virology, 1992. 66(8): p. 5047-5058. 10. Folcik, V.A., et al., Idiopathic pulmonary fibrosis is strongly associated with productive infection by herpesvirus saimiri. Modern Pathology, 2013. advance online publication. 11. Grimm, V. and S.F. Railsback, Individual-based Modeling and Ecology. Princeton Series in Theoretical and Computational Biology, ed. S.A. Levin. 2005, Princeton, New Jersey: Princeton University Press.

Implementing an Agent-Based Model using OpenCL: A Case Study

Gregory J. Davis Center for Research Computing University of Notre Dame, IN, USA

Klaus Kofler DPS Group, Institute for Computer Science University of Innsburck, Austria

Abstract:

The graphics processing unit (GPU) has become an important resource for computational tasks that can be deconstructed into parallelizable operations. Agent-Based Modeling (ABM) can generally be classified as this form of task. We present an OpenCL implementation of an existing ABM used to simulate populations of Anopheles gambiae mosquitoes, an important vector of malaria transmission, to illustrate the potential improvement in execution time GPUs can offer ABMs. Discussed are methods and techniques used to overcome design challenges that can arise when porting ABMs from traditional object-oriented designs to GPU-based designs. The implications for future agent-based software development frameworks are also discussed. Optimal Control of the SugarScape Agent-based Model

Scott Christley1, Matthew Oremland2, Rene Salinas3, Rachael M. Neilan4 and Suzanne Lenhart5

1Department of Surgery, University of Chicago, Chicago, IL 60637, USA 2Department of Mathematics, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, USA 3Department of Mathematical Sciences, Appalachian State University, Boone, NC 28608, USA 4Department of Mathematics and Computer Science, Duquesne University, Pittsburgh, PA 15282, USA 5Department of Mathematics, University of Tennessee, Knoxville, TN 37996, USA

Abstract:

One of the challenges for the analysis of agent-based models is the determination of a strategy, policy or intervention that can produce a desired outcome from the model. For example, an intervention for a biomedical model to go from a disease state to health, a strategy for controlling an invasive species in an ecological model, or a tax policy that promotes growth in an economic model. Optimal control theory is a mathematical optimization technique that can be used to derive such interventions, however it is only applicable to continuous systems as described by differential equations and cannot be directly applied to an agent-based model. We posit that if a differential equation model is designed that approximates the agent-based model, then an optimal control derived for that different equation model may be applicable to control the agent-based model. In this research, we describe how we utilize this idea of model equivalency to apply optimal control theory to the well-known SugarScape agent-based model. The SugarScape model was introduced in Epstein & Axtell’s book, Growing Artificial Societies: Social Science from the Bottom Up. It is a simple agent-based model that shows how wealth inequality can occur among a population of agents with heterogeneous attributes in a heterogeneous environment. Applying optimal control theory to SugarScape requires a series of steps: 1) define an objective function that captures the desired outcome as a set of constraints, 2) approximate SugarScape with a differential equation model, 3) derive the optimal control for the differential equation model and the objective function, 4) transform and apply the optimal control into SugarScape. Lastly, we can evaluate simulation results from applying the control to SugarScape and determine if we achieved the desired outcome. We will describe the challenges and results from each of these steps, and provide some insights into the possibility of using optimal control theory for more sophisticated agent-based models.

A Multi-Level Complex Adaptive System Approach for Modeling of Schools

Ted Carmichael1, Mirsad Hadzikadic2, Mary Jean Blink1, John C. Stamper1,3 1TutorGen, Inc., Wexford, PA, USA {tcarmichael, mjblink}@tutorgen.com 2University of North Carolina at Charlotte, Charlotte, NC, USA [email protected] 3Carnegie Mellon University, Pittsburgh, PA, USA [email protected]

Abstract. The amount of data available to build simulation models of schools is immense, but using these data effectively is difficult. Traditional methods of computer modeling of educational systems often either lack transparency in their implementation, are complex, and often do not natively simulate non- linear systems. In response, we advocate a Complex Adaptive Systems approach towards modeling and data mining. By simulating agent-level attributes rather than system-level attributes, the modeling is inherently transparent, easily adjustable, and facilitates analysis of the system due to the analogous nature of the simulated agents to real-world entities. We explore the design a CAS model of schools using multiple levels of data from varied data streams.

Keywords: Complex Adaptive Systems, Agents, Educational Data Mining

1 Multi-Level CAS Design of an Educational System

As schools become increasingly wired, the ability to collect data at multiple levels has grown exponentially to the point of becoming overwhelming. We classify the multiple data streams into four levels: Individual, Classroom, School, and District. This work is centered on finding the complementary links between these levels and using them together to bring a much clearer picture of the overall educational system. At the highest levels, most of the academic work in the fields of learning analytics, educational data mining, and intelligent tutoring systems focus specifically at the classroom level or the individual student level using data from learning management systems or finer grain data from logs created from educational technologies[3]. Some work has brought together log data and correlated it with student grades, but little has been done to harness all of these data streams into a robust model. We propose a CAS (Complex Adaptive System) model to do this, for two reasons: the inherent transparency of using agent-based analogues, and the ease with which a CAS model can represent non-linearities. Educational systems currently collect many characteristic-, performance- and outcome-level data, including grades, test results, economic status, gender, age, race, etc. However, such data, while useful, still leave many aspects of classroom performance unreported. For example, none of them include the nature and frequency of interactions among students, teachers and students, students and principals, teachers and principals, or principals and superintendents. In addition, there are no correlations between the availability of resources, the nature of such interactions, and the overall performance of students and schools/school districts. Due to the interactive nature of the classroom there is also a great potential for threshold “tipping point” effects to exist, and it is intuitively true that some students or student clusters can have an outsize effect on the rest of the class. One of the goals of this research will be to discover and understand the underlying dynamics of such threshold effects, within the classroom, the school, and the district-wide school system, so that a smarter approach in resource allocation can produce a more effective educational system. This work identifies the links between multiple streams of data and the development of CAS model to represent an entire school ecosystem, from the individual student to the district level. The end result of this effort will produce a robust model of an educational system at multiple scales, one that can not only help determine the causal factors of desirable outcomes, but also allow for multiple “what if” scenarios to be run in simulation, so that these outcomes can be improved and resources are expended in the most efficient manner.

References

1. Carmichael, T., Hadzikadic, M., Dr_eau, D., and Whitmeyer, J. (2009). Towards a General Tool for Studying Threshold Effects Across Diverse Domains Springer-Verlag. pp. 41–62. 2. Gell-Mann, M. (1994). Complex Adaptive Systems. Addison-Wesley. pp. 17–45. 3. Stamper, J., Carmichael, T. (2007). A Complex Adaptive System Approach to Predictive Data Insertion for Missing Student Data. In Proceedings of the 3rd Int. Conference on Computer Blended Learning (ICBL 2007), Florinopolis, Brazil, May 2007. Kassel Press. Crypto-economic Design: A Proposed Agent-Based Modelling Effort

Dave Babbitt, Northwestern University Joel Dietz, University of Pennsylvania

A crypto-economy is an economic system which is 1) not defined by geographic location, political structure or legal system, and 2) uses cryptographic techniques to constrain behaviour (in place of using trusted third parties). Economic agents in these systems can be human-controlled clients and autonomous organizations or contracts. Prices of transacted goods and services in these economies are expressed in a built-in money-like informational commodity (a "crypto-currency") and all transactions are recorded on a public ledger.

They are important because they eliminate "bridging" social capital - the building of connections between heterogeneous groups - as a necessary precondition for successful economic development (Schuller, Baron, & Field, 2000). You no longer have to trust your valuables to strangers.

Crypto-economies have more than a need for software and security testing - they need to have their economies tested. Adding to the complexity of crypto systems (that a crypto-economy is based on) is the fact that exchanging value necessarily involves economic considerations. Therefore they must be analysed not only for computational soundness and security, but also for economic soundness (Poelstra, 2014). That is, they must be designed so that incentives are aligned with the goal of strengthening the security of the system and not inadvertently weakening it.

Agent-Based Modelling (ABM) often results in what is called "weak emergence" - appearance of new properties not fully reducible to that of the micro-properties on which it supervenes, but derivable only by simulation (Bedau, 1997). It is this weak emergence and the relative ease of capturing salient aspects of the actual system that allows development of crypto-economy-specific test scaffolding.

To capture salient aspects of crypto-economic systems we can categorize economic agents into "speculators", "miscreants", and "altruists". And open code sources of crypto-currencies can be written almost character for character into the agents built for the test, adding pauses in place of the cryptographic calculations. The cycle of enthusiasm and strong feeling around the adoption of the new economy can be incorporated as an institutional arrangement.

Speculators, for instance, can be modelled as standard profit-seeking economic agents. The altruists and malicious are harder agents to model. The altruists can be modelled with information cascades, where an agent observes the actions of others and then — despite possible contradictions in its own private information signals — engages in the same acts. As theoretical security holes are discovered, miscreants can be modelled as exploiting them.

Economic analysis of crypto-economies exposes various public goods issues that typically happen to a communal effort in danger of being lobbied by special interest groups. These public goods issues include 1) the financial incentives for operating a centralized mining pool, 2) the centralization of infrastructure without the benefits of centralization (i.e., lower transaction costs, efficiencies of scale), and 3) the lack of financial incentives for working as a developer.

The first economists to study Bitcoin have attempted to discover what individual incentives exist without the use of agent-based models (ABMs). We conclude that these issues would've been readily apparent ("low hanging fruit") with the use of ABMs and that if you run an economic simulation of the design ahead of time, you will at least have a model-based exploration method to find the big issues.

Works Cited

Arrow, K. J., & Debreu, G. (1954). Existence of an equilibrium for a competitive economy. Econometrica: Journal of the Econometric Society, 265-290.

Axelrod, R. M. (1997). The complexity of cooperation: Agent-based models of competition and collaboration: Princeton University Press.

Bedau, M. A. (1997). Weak emergence. Nous, 31(s11), 375-399.

Bonabeau, E. (2002). Agent-based modeling: Methods and techniques for simulating human systems. Proceedings of the National Academy of Sciences of the United States of America, 99(Suppl 3), 7280-7287.

Branwen, G. (2014). BITCOIN IS WORSE IS BETTER. gwern.net.

Demazeau, Y. (1995). From interactions to collective behaviour in agent-based systems. Paper presented at the In: Proceedings of the 1st. European Conference on Cognitive Science. Saint-Malo.

Kai Chang, M. B. a. S. D. (2014). The MtGox 500. Stamen Design. Retrieved May 13, 2014, from http://bitcoin.stamen.com/

Kuznicki, J. (2013). These Three Graphs Prove That Bitcoin Is a Speculative Bubble (Vol. 2014). http://ordinary-gentlemen.com/: Ordinary Times: on Politics and Culture.

Mike, G., Ripper234, et al. (2014). Scalability. Bitcoin wiki. Retrieved April 27, 2014, from https://en.bitcoin.it/wiki/Scalability

Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. Consulted, 1, 2012.

Perry, D. (2012). Measuring Bitcoin Speculation. Coding In My Sleep.

Poelstra, A. (2014). A Treatise on Altcoins. Retrieved from https://download.wpsoftware.net/bitcoin/alts.pdf

Schuller, T., Baron, S., & Field, J. (2000). Social capital: a review and critic. In S. Baron, J. Field, & T. Schuller (Eds.). Social Capital. Oxford: Oxford University Press.

Tesfatsion, L. (2006). Agent-based computational economics: A constructive approach to economic theory. Handbook of computational economics, 2, 831-880. A Spatial Agent-Based Model of Anopheles vagus for Malaria Epidemiology

Md. Zahangir Alam1, S. M. Niaz Arifin2, M. Sohel Rahman1

1Department of Computer Science & Engineering (CSE), Bangladesh University of Engineering & Technology (BUET), Dhaka 1000, Bangladesh 2Department of Computer Science and Engineering, University of Notre Dame, IN 46556, USA

Introduction

Malaria is the ninth largest cause of global human mortality and morbidity [1, 2]. About 3.3 billion people in 99 countries are reported to be at risk of malaria [3]. Each year, it kills around two million people [4], most of which are young children in sub-Saharan Africa [5]. Being a mosquito-borne disease, malaria is transmitted among humans by female mosquitoes of the genus Anopheles [6].

Among the approximately 430 Anopheles species, only 30-40 are known to transmit malaria in nature [6]. Anopheles gambiae is responsible for transmitting the most dangerous malaria parasite, namely, Plasmodium falciparum, among humans. On the other hand, Anopheles vagus is another species that transmits Plasmodium vivax, another dangerous parasite causing 47% malaria cases in the Asia-Pacific Region [7, 8]. An. vagus is widely distributed in Asia, particularly in Bangladesh, Cambodia, China (including Hong Kong), India, Indonesia, Laos, Malaysia, Mariana Islands, Myanmar, Nepal, Philippines, Sri Lanka, Thailand, and Vietnam [10-12].

In the recent past, several mathematical (equation-based) and agent-based models (ABMs) of for malaria have been developed to model the life cycle of An. gambiae [13-19, 22, 23]. The spatial dimension, using a landscape-based approach, is also described in detail by some of the models [14, 15].

However, despite its wide distribution in the Asia-Pacific Region, no model of An. vagus has yet been developed or reported in the literature. In this paper, we describe the design, implementation, and some preliminary results from of an ABM of for An. vagus. The ABM, denoted as ABMvagus, is developed by modifying an established existing ABM of for An. gambiae (referred to as ABMgambiae henceforth) from the University of Notre Dame [13-16]. We describe the life cycle modeling of An. vagus, based on its important biological parameters, and report the effect of temperature on the abundance of An. vagus. Obtaining monthly An. vagus female abundance data from field studies [10], we validate the model’s output against the real data.

Model Development

Like ABMgambiae, ABMvagus includes two distinct phases in the An. vagus life cycle, namely, aquatic and adult. The aquatic phase consists of three stages, namely, egg, larva and pupa. The adult phase consists of five stages, namely, immature adult, mate seeking, blood meal seeking, blood meal digesting, and gravid. The major differences between the two ABMs are reported in Table 1.

Both An. vagus and An. gambiae pass through the same stages during their life cycle [6]. However, An. vagus mostly rests indoors [20]. ABMvagus primarily considers the eight stages, and modifies/extends ABMgambiae according to recent field data for each stage.

Table 1: Differences between the ABMs of Anopheles vagus and Anopheles gambiae. DMR denotes the daily mortality rate.

Model Feature An. vagus An. gambiae Reference Reference Egg 60% eggs are developed [25,26] Development [13] development within 2 days and remaining (incubation and and DMR 40% within 3 days in normal hatching) is temperature. DMR is 10%. temperature dependent, equation-based. DMR is 10%. Larval Four sub-stages: 1st instar, 2nd [25,26] Development is [13] development instar, 3rd instar & 4th instar temperature dependent. and DMR with different duration. DMRs in each sub-stage are 15%, 10%, 10%, and 10%, respectively. Pupa 40% pupae are developed [25,26] Development is [13] development within the first 24 hours and temperature dependent, and DMR 60% within the next 30 hours. equation-based. DMR is 5%. DMR is 10% Immature Adult 10% emergence on the 6th day, [25,26] Development is [13] 10% on the 7th day, 40% on temperature dependent, the 8th day, 30% on the 9th equation-based. day, and 10% on the 10th day. DMR is 10% Blood Meal Continues until it gets blood 24 Continues until it gets [13] Seeking meal or dies. The most blood meal or dies. The effective time-window for effective time-window host-seeking is 5.00am to for host-seeking is 6.00am. 6.00pm to 6.00am. Host-seeking occurs as follows: at 8:30 pm: 0%, 9:30 pm: 13.67%, 10:30 pm: 15.83%, 11:30 pm: 10.8%, 12:30 am: 7.2%, 1:30 am: 0.72%, 2:30 am: 0%, 3:30 am: 0.72%, 4:30 am: 1.44%, 5:30 am: 35.25%, 6:30 am: 14.39%, 7:30 am: 0%.

In the model, mosquitoes and aquatic habitats are modeled as agents. A mosquito agent stays in each stage for certain duration, with probabilistic transitions to the next stage. For example, an agent oviposits 60% of the eggs on the 2nd day, and the remaining 40% eggs on the 3rd day in normal temperature (i.e, 26- 30°C). For each stage, the daily mortality rate (DMR), obtained from field data, is applied after it is converted to hourly mortality rates (for each simulated hourly timestep) to match ABMgambiae.

P. vivax malaria transmission depends on several factors, including vector availability, biting rates, etc. Many of these factors are also influenced by weather and climate variables, especially temperature [11]. For a particular geographic region, daily temperature primarily affects larval development and blood meal digesting durations. Hence, ABMvagus includes a temperature profile module, in which annual temperature data is loaded, and the simulations are fed with daily temperature data derived from the profile.

Field Data Collection

Field data on An. vagus abundance are collected from a study in the hill tract district of Bandarban, Bangladesh, which reports abundances of several local species as follows: An. jeyporiensis: 18.9%, An. vagus: 16.8%, and An. kochi: 14.4% [10]. Monthly An. vagus female abundance is reported to reach the highest level during March, followed by an immediate sharp decrease during April. Daily temperature data is also collected from the Soil Resource Development Institute for Bandarban, Bangladesh [21].

Verification & Validation

The conceptual model is verified through early testing to check whether the implementation is a correct realization of the concepts adopted from the field data. Two early implementations are compared with each other: one with twelve stages in which the Larva stage is further sub-divided into four sub-stages (see Table 1), and the other with eight stages where larval development is calculated with a temperature- dependent equation. Results from both implementations are compared in order to select the more correct one (in terms of realization of the conceptual model). The model is also validated against field data (as described above) [10, 21].

Results

Some preliminary results derived from ABMvagus are shown in Figures 1 and 2.

Figure 1. Female Abundance in 4 years simulation run: Female Abundance (FA) from a 4-years simulation run. The annual pattern of An. vagus abundance is directly regulated by temperature. The x- axis denotes simulation time (in days) and the y-axis denotes mosquito abundance.

Figure 2: Model Validation. An. vagus abundances from the simulations of three consecutive years are compared to field data, showing that the simulated results are very close to field data. This helps to ensure the validity of the ABM.

References

1. Welcome Trust, Malaria Atlas Project (2010). Available from http://www.map.ox.ac.uk 2. WHO, Global burden of disease (2008). http://www.who.int/healthinfo/global_burden_disease/en/ 3. WHO, Larval source management – a supplementary measure for malaria vector control. An operational manual (July 2013). http://www.who.int/malaria/publications/atoz/9789241505604/en/ 4. CDC (Centers for Disease Control and Prevention), Malaria Facts. http://www.cdc.gov/malaria/facts.htm 5. Snow R.W., Guerra C.A., Noor A.M., Myint H.Y., Hay S.I.: “The global distribution of clinical episodes of Plasmodium falciparum malaria.”, Nature, 343 (7030): 214-7, 2005 6. CDC (Centers for Disease Control and Prevention), Anopheles Mosquitoes. http://www.cdc.gov/malaria/biology/mosquito 7. Maheshwary N.P., Majumdar S., Chowdhury A.R.,Fruque M.S., Montanari R.M.: “Incrmination of Anopheles vagus Donitz, 1902 as an Epidemic Malaria Vector in Bangladesh. Indian Journal of Malariology”, Vol. 31, March 1994, pp. 35-38. 8. Lynch C., Hewitt S.: “Malaria in the Asia-Pacific: Burden, success and challenges”, October 2012. 9. Gu W. and Novak R. J.: “Agent-based modelling of mosquito foraging behavior for malaria control”, Transactions of the Royal Society of Tropical Medicine and Hygiene, 103(11):1105- 1112, 2009. 10. Alam M.S., Chakma S., Khan W.A., Glass G.E., Mohon A.N., Elahi R., Norris L.C., Podder M.P., Ahmed S., Haque R., Sack D.A., Sullivan D.J., Norris D.E.: “Diversity of anopheline species and their Plasmodium infection status in rural Bandarban, Bangladesh”, Parasites & Vectors 2012, 5:150 11. Wardrop N.A, Barnett A.G., Atkinson J. and Clements A.C.: “Plasmodium vivax malaria incidence over time and its association with temperature and rainfall in four counties of Yunnan Province, China”, Malaria Journal 2013, 12:452 12. Rueda, L.M., Pecor, J.E. and Harrison, B.A.: “Updated distribution records for Anopheles vagus (Diptera: Culicidae) in the Republic of Philippines, and considerations regarding its secondary vector roles in Southeast Asia”, Tropical Biomedicine 28(1): 181–187 (2011) 13. Zhou Y., Arifin S.M.N., Genetile J., Kurtz S.J., Davis G.J., Wendelberger B.A, Madey G.: “An Agent-based Model of the Anopheles gambiae Mosquito Life Cycle”, SCSC '10 Proceedings of the 2010 Summer Computer Simulation Conference, Pages 201-208, 2010-07-11 (yyyy-mm-dd) 14. Arifin S.M.N., Davis G. K., Zhou Y.: “Modeling Space in an Agent-Based Model of Malaria: Comparison between Non-spatial and Spatial Models”, ADS '11 Proceedings of the 2011 Workshop on Agent-Directed Simulation, Pages 92-99, 2011-04-03 (yyyy-mm-dd) 15. Arifin S.M.N., Davis G.J., Zhou Y.: “A Spatial Agent-Based Model of Malaria: Model Verification and Effects on Spatial Heterogeneity”, International Journal of Agent Technologies and Systesm, 3(3), 17-34, July-September 2011 16. Arifin S.M.N., Madey G.R., Collins F.H.: “Examining the impact of larval source management and insecticide-treated nets using a spatial agent-based model of Anopheles gambiae and a landscape generator tool”, Malaria Journal 2013, 12:290 17. Macdonald G.:“The epidemiology and control of malaria.” Oxford university Press, London, 1957 18. Dietz K., Molineaus L. and Thomas A.: “A malaria model tested in the African savannah” Bulletin of the World Health Organization 50, 347-357,1974 19. Smith T, Maire N, Ross A, Penny M, Chitnis N, Schapira A., Studer A., Genton G., Lengeler C., Tediosi F., Savigny D.D. and Tanner M. “Towards a comprehensive simulation model of malaria epidemiology and control”, Parasitol, 2008 20. Nagpal B.N., Sharma V.P., “INDIAN ANOPHELES”, Science Publishers, Inc.; 1995 21. Soil Resource Development Institute (SRDI), Bandarban, Bangladesh. http://www.srdi.gov.bd/ 22. Gu W. and Novak R. J.: “Agent-based modelling of mosquito foraging behavior for malaria control”, Transactions of the Royal Society of Tropical Medicine and Hygiene, 103(11):1105- 1112, 2009. 23. Eckhoff P.: “A malaria transmission-directed model of mosquito life cycle and ecology”, Malar J 2011, 10:303. 24. Quraishi Sayeed M.: “Nocturnal Prevalence of Anopheline Mosquitoes in Mymensingh District, East Pakistan1”, Journal of Economic Entomology, Volume 56, Number 5, October 1963, pp. 670-672(3) 25. Alam Mohammad Shafiul, International Centre for Diarrhoeal Disease Research Bangladesh (icddr,b), Personal Communication. 26. Al-Amin H M, International Centre for Diarrhoeal Disease Research Bangladesh (icddr,b), Personal Communication. Agent-Based Microsimulation (ABµS) Modeling: Revisiting the Micro Perspective

S. M. Niaz Arifin(a), Rumana Reaz Arifin(b), Gregory R. Madey(c) (a, c)Department of Computer Science and Engineering, University of Notre Dame (b)Department of Civil & Environmental Engineering & Earth Sciences, University of Notre Dame

(a)[email protected], (b)[email protected], (c)[email protected]

Microsimulation models (MSMs) fall under a category of computerized analytical simulation models that can perform highly detailed analysis of activities. Introduced in the late 1950s by Guy Orcutt, MSMs are suitable to model interactions between the design and implementation of policies and individual decision making units. Frequently, MSMs involve the generation of data on social or economic units (e.g., persons, households, or firms) drawn from survey-based microdata. MSMs enable us to examine the impact of policy changes on individual decision units, and this micro-level focus distinguishes them from other modeling paradigms. In contrast, agent-based models (ABMs) and cellular automata (CA) became increasingly popular as modeling approaches in the social sciences because of their ability to directly model individual entities and their interactions.

Although both MSMs and ABMs have been successfully used in the recent past to model complex social (and other types of) systems, there are significant variations of emphasis between the two approaches. MSMs, having a strong applied policy focus, usually are not suitable to model the behaviors of individuals, interactions between individuals, heterogeneous populations, learning, emergence, or other types of adaptive features, etc., for which ABMs are usually better suited. In addition, in many simulation studies, the analysis of spatial and temporal dimensions bears special importance. Since geography usually has an important impact on human activities, often modeling the social system with its local context seems important and necessary. Thus, new paradigms have also been proposed that integrate the power of geographic information systems (GISs) with ABMs. It has also been argued that in recent hybrid models, a GIS can provide a bridge to link MSMs and ABMs.

In order to combine the best features of MSMs and ABMs, several studies have attempted to use hybrid, unified approaches that can offer a synthesis of the three paradigms. Such approaches may deliver higher potential in the advance of the simulation and modeling methodology. In this paper, we formally establish the notion of a hybrid, unified approach which we call agent-based microsimulation. To distinguish it from agent-based modeling & simulation, ABMS, we denote it by ABµS, where the “µ” stands for the “micro” prefix. Extending an earlier study that integrates an ABM of malaria epidemiology with a GIS, we argue that detailed analysis of the micro perspective can offer additional capabilities and benefits that are often overlooked by traditional ABMs.

The proposed ABµS methodology is applied to model malaria, which is one of the oldest and deadliest infectious diseases in humans, transmitted by female mosquitoes of the genus Anopheles. Unfortunately, most malaria ABMs, despite modeling individual agents and their interactions, still simulate a macro system. Although they possess the ability to explore a multitude of new insights by tracking each individual agent, almost none takes any advantage of it, and most focus on the aggregate behavior of agents. The outputs from these models are thus mostly restricted to the macro-level impacts on the system, such as time series plots that depict the proportional changes of various disease parameters (e.g., incidence, prevalence, entomological inoculation rate (EIR), etc.). Thus, the powerful insights that can be explored by modeling and simulating the actions and interactions of the smaller scale units (agents) are often overlooked.

In addition to the macro-level analysis, the ABµS methodology will permit to analyze the micro perspective in the local geographical context for a selected region. Many well-conceived field- based studies of malaria have shown that due to the complex nature of the disease, vector control interventions such as insecticide-treated nets (ITNs) and indoor residual spraying (IRS) may have subtle impacts on the mosquito and human populations. Traditional macro-level outputs of ABMs are often inadequate to capture these. However, ABµS permits the modeling of these features, as illustrated below with two examples drawn from the literature.

● Community effect of ITNs: Studies from various parts of Africa showed that with sufficient coverage of ITNs, there is an overall suppression of the mosquito population, resulting in a mass community effect of the insecticide that reduces malaria transmission. With the micro-level focus of ABµS, modeling and analysis of such secondary, subtle effects can be easily performed.

● Policy formulation with limited resources: While simulating the impact of an intervention on a population, often the resources appear to be limited. For example, in a densely populated area with limited number of bednets, traditional ABMs may not be suitable to devise the best resource allocation plan. In such cases, ABµS can simulate and quantitatively analyze the best allocation and use of the limited resources, thus assisting in future policy formulation and decision making by public health officials.

Future Work

The current work is in progress. We plan to apply the proposed ABµS methodology to specific geographic regions where the disease is endemic, and different levels of transmission occur throughout the year. An Initial Agent Based Model for! Innovation Ecosystems! Mustafa Ilhan Akbas, Ivan Garibay University of Central Florida ! Email: {miakbas, igaribay}@ucf.edu The global economies have been trying various methods of investments in innovation to raise productivity, create jobs and improve life standards. For the same purpose, the United States has long been supporting the foundation of basic research and innovation for technological advances, which generate wealth over time. The productivity gains through technological advancements, labor specialization and innovation have been ar- ticulated by evolutionary economists since the middle of the twentieth century [1, 2]. However, it is still open to research how certain innovation ecosystems such as Silicon Valley are extremely productive and grow continuously while other similar systems lan- guish. Therefore it is critical for the national economic well-being to study, understand !and more efficiently create the innovation ecosystems. The economic entities in an innovation ecosystem are intertwined such that the success of an innovation depends not only on the innovating entities, but also on the suppliers and consumers of those entities. All of the members of an ecosystem coevolve and the innovations of an entity result in the innovations of others [3]. Over time, the technology space of the ecosystem changes in response to the innovations. In order to reflect these characteristics, we aim to model the innovation ecosystems as non-linear, complex adaptive networks in which economic agents are connected by social networks and compete, cooperate and adapt to each other’s needs forming unplanned consequences in the network. The goal of creating such a model is to improve the understanding of in- novation ecosystem dynamics using agent-based computational economics to inform !policy makers and test policy hypotheses. In this paper, we first explore the state of the art agent-based innovation ecosystem models and work on the compartmentalization of these models. In particular, we classify the models in terms of their simulation methods, objectives and approaches in the adop- tion of stochastic processes [4, 5]. Then we present our bottom-up approach to create our model and conduct simplistic experiments, whose purpose is to illustrate the initial ideas behind the model framework, rather than to accurately describe the whole innova- tion ecosystem. As we focus on a system in which the agents interact and compete, we study the basic minimal ecosystem dynamics of our model by comparing its behavior to systems with trophic functions [6, 7]. Overall, this study 1) demonstrates the critical points in innovation ecosystem models including current theoretical and methodological contributions; and 2) identifies the position of our approach in the classification of mod- !els and studies its minimal ecosystem dynamics. ! References [1] Schumpeter, Joseph A. Capitalism,socialism, and democracy. Harper Perennial Modern Classics, 2008. [2] Heilbroner, Robert L. The worldly philosophers: The lives, times and ideas of the great economic thinkers. Simon and Schuster, 2011. [3] Adner, R., & Kapoor, R. Value creation in innovation ecosystems: How the structure of technological interdependence affects firm performance in new technology genera- tions. Strategic management journal, 31(3), 306-333, 2010. [4] Farmer, J. D., and Foley, D. The economy needs agent-based modelling. Nature, 460(7256), 685-686, 2009. [5] Gatti, D., Desiderio, S., Gaffeo, E., Cirillo, P., and Gallegati, M. Macroeconomics from the Bottom-Up. New Economic Windows. Springer, 2011. [6] Lotka, A.J., "Contribution to the Theory of Periodic Reaction", J. Phys. Chem., 14 (3), pp 271–274, 1910. [7] Gandolfo, G., "Giuseppe Palomba and the Lotka–Volterra equations", Rendiconti Lincei, 19(4), 347–257, 2008. An Agent Based Modeling Approach to Predicting the Effect Anthropogenic Pressures on the Movement Patterns of Mongolian Gazelles

Connor Gibb, Michael Kleyman, Maria Koebel, Rebecca Natoli, Kyle Orlando, Matthew Rice, Claire Weber, Will Weston-Dawkes, Bill Fagan

University of Maryland

Mongolian Gazelles are an ungulate species inhabiting the Eastern Mongolian Steppe. Due to the highly unpredictable and dynamically heterogeneous nature of that landscape, Mongolian gazelles are highly nomadic foragers; they roam across a large range in the landscape and their movements do not follow a predetermined migratory path. They also regularly split from and join herds of other gazelles that they encounter. Our research project aims to create a computer model of Mongolian Gazelle movement in response to anthropogenic changes in the landscape. The nomadic nature of the gazelles, as well as the highly variable landscape they live on, makes creating a model of their movements a complex task at the population level. To handle this problem we used an Individual Based Neural Network Genetic Algorithm (ING) model. We model the movement of these nomadic animals at the individual level, allowing population-level movements to emerge naturally as a result of many individual gazelle movement decisions. An individual’s movement decisions are generated by a neural network, a computational representation of the central nervous system, implemented as a weighted graph. Each artificial gazelle has a list of genes which corresponds to weights on the neural network. The weights are optimized using a genetic algorithm, which mimics the biological processes of crossing over, mutation, and selection based on fitness. Each artificial gazelle has a fitness score that is determined by how similar its movements are to those observed in real gazelles. We define real gazelle movements using five statistical metrics, which together represent gazelle movement: displacement vs. time lag, population dispersion index (the gazelles’ proximity to one another), movement coordination index (how similar one gazelle’s movements are to the movements of those around it), frequency of gazelle time points vs. distances from human-created landscape features, and frequency of time spent on grass patches as a function of the grass patch’s NDVI value. We calculated these metrics from real gazelles using pre-obtained data of gazelle movements from satellite collars. We also have NDVI data, a satellite-derived measure corresponding to vegetation greenness and the amount of grass in an area. We used the NDVI data to create a series of landscapes corresponding to actual change in vegetation over time. In addition, we used data on the location of human-populated areas and human-created landscape features to generate anthropogenic features within our model. The fitness of an artificial gazelle will determine how likely it is to contribute to the next generation of artificial gazelles. The genetic algorithm also uses artificial mutation and crossing over to add variation to the ‘genes’ of the artificial gazelles. We ran our model for many generations until we obtained a population of artificial gazelles which have nearly identical movement behaviors to those observed in real gazelles as determined by our statistical metrics. We infer that such a close match means that the population of artificial gazelles accurately mimics real gazelle movement decisions. We ran the trained model with different scenarios of anthropogenic features on the landscape to see what will happen to the gazelle population over time. We extracted from the model information on what landscape features would be harmful or even devastating to the survival of the gazelle population, a task that was previously rendered impossible by the nature of the gazelle’s dynamic movements. We implemented our model using the Repast Simphony Agent Modeling Framework. The Repast framework supports three languages: ReLogo, Groovy, and Java. We chose Java for its performance benefits. Repast also provides visualization capabilities to agent based models. In addition, it provides a continuous landscape and a synchronous event scheduler for our model. Using the Repast framework, one can see the movements of our artificial gazelles overlayed on the landscape over time. Identifying which anthropogenic pressures would restrict Mongolian gazelle movement and drastically decrease population size can help with future landscape planning efforts in the Mongolian Eastern Steppe. We hope other researchers will able to use our model to understand movements of other ungulate species in relation to anthropogenic pressures.

THE IMPACT OF ANTIBODY DEPENDENT ENHANCEMENT ON DISEASE DEMOGRAPHICS AND TRANSMISSION POTENTIAL OF MULTI-SEROTYPE INFECTIOUS DISEASES

Q.A. TEN BOSCH, B.K. SINGH AND E. MICHAEL UNIVERSITY OF NOTRE DAME, NOTRE DAME, INDIANA, USA

INTRODUCTION The impact of serotype interactions on the epidemiological dynamics of multi- serotype pathogens has been studied extensively, in particular for dengue virus (DENV). Dengue is a mosquito borne viral disease, which causes an increasingly high burden in the tropics and subtropics. Dengue dynamics are characterized by irregular annual and multi-annual cycles and complex patterns of serotype replacement. These are believed to arise from the interplay between environmental factors, serotype interactions and predator-prey dynamics. Antibody dependent enhancement (ADE) is such a serotype interaction and manifests itself in an increased susceptibility to heterologous serotypes after individuals have recovered from a primary infection. This results in a selective advantage of less prevalent serotypes, a consequent replacement of the common serotype and presumably the irregular cycle of serotype replacement as is observed in many multi-serotype pathogens. Mathematical models have been used to disentangle the role of ADE and other intrinsic and extrinsic drivers of the dynamics of dengue. The commonly used deterministic models rely on the inclusion of complexities such as ADE to mimic the characteristic chaotic oscillations, whereas their agent-based alternatives can replicate such dynamics under simpler assumptions upon the inclusion of stochasticity. To tease out which behaviors arise from model choice and which are indeed inherent to the disease, we examine whether findings from deterministic models hold in an agent-based framework, with a specific focus on the effect of ADE.

METHODS To study the impact of ADE on multi-serotype disease dynamics, we constructed a two-serotype agent-based transmission model under the assumptions of homogeneous mixing and direct transmission, i.e. the mosquito population is not explicitly modeled. Individuals move around randomly in one of 8 states: Individuals are borne fully susceptible to both serotypes. Upon an encounter with an infectious individual, they get infected with serotype 1 or 2 with probability p. Individuals recovered from their first infection remain immune to that serotype but experience enhanced susceptibility to the other serotype as a result of ADE. After recovery from a secondary infection, the individual is assumed to hold life-long immunity against both serotypes. No access disease mortality is assumed, leaving individuals to die solely of old age. Birth processes are regulated by the environment’s carrying capacity. Both infection and recovery are assumed to be stochastic processes. RESULTS In conjunction with earlier research, the two-serotype agent-based transmission model demonstrates oscillations of the serotypes, also when no ADE is assumed. The inclusion of ADE results in a moderate increase in the force of infection, defined as the average number of new infections per infected individual per time step. This is in contrast with analytical findings that the transmission potential is independent of ADE. Additionally, we found that an increased level of ADE (defined as the rate of increase of susceptibility upon a primary infection) results in an increase in the mean age of primary infection, yet a decrease in the age of acquiring a secondary infection.

DISCUSSION Cross-verification between compartmental and agent-based models is important, in particular to distinguish disease characteristics from model artifacts. The inclusion of stochastic, individual-based disease dynamics to our model introduces oscillations not observed in deterministic models of similar simplicity. Consequently, estimates of the level of ADE required to mimic the irregular fluctuations observed in dengue case data may be overestimated by deterministic models, this as a result of simplifying assumptions on natural variability in the transmission process. Additionally, the use of an agent-based model in this context allows us to derive measures not easily attainable from deterministic models. For instance, the agent-based character of the model enables us to derive the reproduction number directly from the simulations and thereby test analytical findings. While ADE may not affect the threshold behavior of the model, this model does demonstrate a positive effect on the transmission potential. Additionally, the agent-based approach permits one to track the mean age of infection. This can aid in future endeavors such as explaining differences in the mean age of dengue cases across different endemic dengue regions, as well as investigating the increase in the mean age of dengue onset as observed over the last decades in Thailand.

Application of microsimulation modeling for malaria control decision-making

Stuckey, EM1,2

1Swiss Tropical and Public Health Institute, Basel, Switzerland 2University of Basel, Basel, Switzerland

Background Financing for malaria control has increased substantially over the past decade, funding large scale distribution of malaria control commodities such as long-lasting insecticide-treated nets (LLINs) and artimisinin combination therapy (ACTs). Over the same time period there has been a noted decrease in reported morbidity and mortality due to malaria. According to the 2012 World Malaria Report, 50% of malaria endemic countries were on track to achieve the goal of 75% reduction in malaria cases by 2015 [1]. However, this success is being challenged by the lack of sufficient funding for malaria control. Roll Back Malaria estimated a gap of $3.8 billion dollars to fund sufficient malaria commodities over only the period of 2013-2015 [2]. This context justifies the prioritization of identifying and assessing the cost-effectiveness of different mixes of malaria control interventions.

Methods The goal of this work is to apply individual-based stochastic models of malaria to field sites to better understand transmission dynamics and explore different control interventions and strategies. Simulations were conducted using OpenMalaria, an ensemble of stochastic simulation models of malaria transmission able to simulate the dynamics of Plasmodium falciparum in a given population [3]. Based on parasite densities for individual infections, stochastic individual- based models of malaria in humans [4,5] are linked to a periodically-forced model of malaria in mosquitoes [6] in order to simulate the dynamics of malaria transmission and the impact of intervention strategies for malaria control. Models are fitted to 10 objectives using 61 standard scenarios, calibrated by the seasonal pattern of infectious bites per person per year, and run for one human life span to induce a stable level of immunity in the population. The two study areas include Rachuonyo South District, western Kenya and Southern Province, Zambia. For both study areas, baseline scenarios were parameterized and experiments designed in collaboration with research and implementing partners: the London School of Hygiene and Tropical Medicine, the Centers for Disease Control and Prevention, and the Kenya Medical Research Institute in Rachuonyo South [7,8], and the Zambia National Malaria Control Centre and PATH/MACEPA in Southern Province [9]. Baseline scenarios were validated by comparing simulation results with observed data collected in the study area. Experiment results were evaluated by comparing simulated results to the simulated baseline scenario.

Results The experiment set in Rachuonyo South attempted to answer the question of whether there are alternative malaria control strategies that could have a larger impact malaria burden in Rachuonyo South compared to the currently-implemented strategy. Simulation results suggest that while an intervention with long lasting insecticide treated net (LLIN) use by 80% of the population, 90% of households covered by indoor residual spraying (IRS) with deployment starting in April, and intermittent screen and treat (IST) of school children using Coartem® with 80% coverage twice per term had the greatest simulated health impact, the current malaria control strategy in the study area of LLIN use of 56% and IRS coverage of 70% was the most cost effective at reducing DALYs over a five year period. For Southern Province, answering the question of which factors are likely to increase the effectiveness of a mass test and treat (MTAT) campaign, simulation results suggest that the most important determinant of success in reducing prevalence is the coverage of the population achieved by the campaign. However, even with high coverage of mass drug administration (MDA) in areas with a pre-intervention all-age parasite prevalence of less than 10%, simulations suggest that elimination would require more than one year of campaign implementation. Including single low-dose primaquine, which acts as a gametocide, to the drug regimen did not further reduce prevalence. The addition of an endectocide, such as ivermectin, resulted in a lower simulated parasite prevalence and warrants further investigation.

Discussion In order to increase the applicability of results and the success of the collaboration, it is essential to have the appropriate use of models to answer a question. Cost-effectiveness analysis is a helpful tool but cannot be the only basis for decision-making; this should take into account logistical feasibility, insecticide and drug resistance, and acceptability of an intervention by the community. For these projects, collaboration between the field and modelers has been essential yet extremely ad hoc. Asking questions that are both epidemiologically relevant and able to be addressed by models requires connections between modelers and the users, who currently are principally funders and academics involved in trial design rather than malaria control program managers. A structure to facilitate these connections does not yet exist, and there is a lack of clarity on who should drive the agenda for questions being asked. Cultivating a broader understanding of the role of models is necessary in order to increase their use in evidence-based decision-making.

Conclusions OpenMalaria has been demonstrated to aid in the decision-making process for trial design and intervention evaluation when applied to operationally-feasible contexts. Based on this application, it can be concluded that a major role of models is to act as a tool to communicate the interactions between elements of a system, and apply them to specific questions. Results of intervention effectiveness are setting-dependent, and models can play a role in bridging the large gap between global predictions and site-specific recommendations. This role and associated use cases is what should in part drive further development of model features and tools

References 1. WHO (2013) World Malaria Report 2012. Geneva: World Health Organization. 2. RBM (2013) Roll Back Malaria Annual Report 2012. Roll Back Malaria Partnership. 3. OpenMalaria (2010) OpenMalaria. 4. Smith T, Maire N, Ross A, Penny M, Chitnis N, et al. (2008) Towards a comprehensive simulation model of malaria epidemiology and control. Parasitology 135: 1507-1516. 5. Smith T, Ross A, Maire N, Chitnis N, Studer A, et al. (2012) Ensemble modeling of the likely public health impact of a pre-erythrocytic malaria vaccine. PLoS medicine 9: e1001157. 6. Chitnis N, Hardy D, Smith T (2012) A Periodically-Forced Mathematical Model for the Seasonal Dynamics of Malaria in Mosquitoes. Bulletin of mathematical biology 74(5): 1098–1124. 7. Stuckey EM, Stevenson J, Cooke M, Owaga C, Marube E, et al. (2012) Simulation of malaria epidemiology and control in the highlands of western Kenya. Malar J 11. 8. Stuckey EM, Stevenson J, Galactionova E, Baidjoe AY, Bousema T, et al. (submitted 2014) Modeling the cost effectiveness of malaria control interventions in the highlands of western Kenya. 9. Stuckey EM, Miller J, Littrell M, Chitnis N, Steketee R (2013) Modeling the Effects of Mass Screen and Treat and Mass Drug Administration Campaigns in Interrupting Malaria Transmission in Southern Province, Zambia. Swiss Tropical and Public Health Institute. pp. 1-22.

SwarmFest 2014

Poster Abstracts

A Spatial Agent-Based Model of Anopheles vagus for Malaria Epidemiology

Authors Md. Zahangir Alam, S. M. Niaz Arifin, and M. Sohel Rahman Bangladesh University of Engineering & Technology, Bangladesh and University of Notre Dame

Abstract

Malaria is the ninth largest cause of global human mortality and morbidity, transmitted among humans by female mosquitoes of the genus Anopheles. Anopheles gambiae is responsible for transmitting the most dangerous malaria parasite Plasmodium falciparum. Anopheles vagus is another species that transmits Plasmodium vivax, another dangerous parasite causing 47% malaria cases in the Asia-Pacific Region. An. vagus is widely distributed in Asia, particularly in Bangladesh, Cambodia, China (including Hong Kong), India, Indonesia, Laos, Malaysia, Mariana Islands, Myanmar, Nepal, Philippines, Sri Lanka, Thailand, and Vietnam. We describe the design, implementation, and some preliminary results from a spatial agent-based model (ABM) of malaria epidemiology for An. vagus. The ABM is developed by modifying an established existing ABM of An. gambiae from the University of Notre Dame. We describe the life cycle modeling of An. vagus, based on its important biological parameters, and report the effect of temperature on An. vagus abundance. Obtaining monthly An. vagus female abundance data from field studies, we also validate the model’s output against real world data. Two early implementations are compared: one with twelve stages in which the larva stage is further sub-divided into four sub-stages, and the other with eight stages where larval development is governed by daily temperature. Genetic-level Modeling of Directed Yeast Evolution in Turbidostats and Chemostats Alexander Madey and Holly Goodson, Dept. of Chemistry and Biochemistry, University of Notre Dame, Notre Dame IN 46556 USA Microorganisms are commonly grown in continuous culture systems called chemostats, which typically have fixed volume and flow rate, so that population size and growth stage are set by limiting nutrition. In a less common type of continuous culture system called a turbidostat, feedback between culture density and flow rate is adjusted so that population is not nutrition-limited and can grow more freely. The goal of this research was to develop a computer model to simulate evolution as it occurs in a turbidostat as compared to a chemostat. A MatLab computer model was developed that uses a simple set of 3 genes with 5 positions each. Initially, we hypothesized that yeast population diversity would be greater in a turbidostat because more genetic combinations could be successful without nutrition limitation (cells in a turbidostat are grown under less selective pressure). Preliminary results analyzed with a genetic diversity algorithm were consistent with this hypothesis. However, additional runs of the program under different conditions suggested that the different population turnover rates in the systems were the primary predictor of population diversity and changes in fitness. Ongoing research focuses on modeling interactions between multiple genes, and working with different mutation rates.

Computational modeling of bacterial motility and social behavior

Authors

Aboutaleb Amiri, Shant M. Mahserejian, Cameron W. Harvey, Morgen E. Anyan, Joshua D. Shrout*, and Mark Alber

Department of Applied and Computational Mathematics and Statistics (ACMS) Department of Civil and Environmental Engineering and Earth Sciences*

Abstract

Pseudmonas aeruginosa is a bacterium that survives in many different environments including the human body where it can cause lung, eye, skin, or gut infections. P. aeruginosa cells have two types of motility appendages: type IV pili (TFP) and flagella. The flagellum at the lagging pole is responsible for their self-propulsion during motility known as swarming. The role of P. aeruginosa TFP during swarming is still unknown. Based on experimental observations, we have developed a computational model to simulate the interactions between cells TFP during swarming. Using this model, we test a proposed mechanism of TFP-TFP interactions within populations of wild type and TFP deficient mutants. Our results show that TFP deficient cells are able to travel through the population more efficiently than wild type cells. We also have studied the social behavior of the bacterium Myxococcus xanthus. Experimental observations suggest that M. xanthus bacteria share certain outer membrane proteins when they are in physical contact. Some of these proteins help bacteria to coordinate group behavior under nutrient limiting conditions while others are needed for cell motility. A combination of computer simulations and cell tracking from experimental data is used to show how this protein sharing between cells impacts their motility and how it depends on bacterial social behavior. We find that in populations with more flexible cells and weaker cell- cell adhesion, proteins are shared more efficiently resulting in faster spread of information. Decentralized K-Means Clustering: Emergent Computation

R. Ryan McCune & Greg Madey University of Notre Dame Computer Science & Engineering Department Notre Dame, IN 46556

A swarm intelligent system is robust, scalable, adaptable, and can efficiently solve complex problems, all through simple behavior. Inspired by biology, swarm intelligent systems, or swarms, utilize emergence, where simple local behaviors distributed across many agents lead to global phenomena, yielding a whole greater than the sum of parts. But the absence of models that quantify emergence, or the lack of an emergent calculus, has challenged swarm engineering. How simple behaviors and interactions lead to complex phenomena is not well understood, let alone developing such behaviors for problem solving. A swarm intelligent solution is presented to a computationally challenging problem with quantifiable results in support of future models of emergence. The swarm intelligent Decentralized K- Means Clustering technique is introduced within the context of rechargeable Mobile Ad hoc Networks (MANETs). Through engineered emergent behavior, cluster centroids relocate to minimize the sum of the squared error between sensors and the nearest centroid, similar to K-means clustering. An agent-based simulation is developed to evaluate the technique, demonstrating the sum of squared error is consistently reduced for both supervised and random scenarios. Lessons Learned from an Experiment in Crowdsourcing Complex Citizen Engineering Tasks with Amazon Mechanical Turk

Presenter: Matthew Staffelbach2

Authors: Dr. Tracy Kijewski Correa1, Dr. Greg Madey2, Dr. Zhi Zhai2, Dr. David Hachen3, Peter Sempolinski2, Dr. Daniel Wei1, Dr. Ahsan Kareem1, and Matthew Staffelbach2

Department of Computer Science and Engineering2 Department of Sociology3 Department of Civil and Environmental Engineering and Earth Sciences1 University of Notre Dame

Abstract

America’s dated infrastructure is failing to keep pace with its burgeoning population. In fact, the average grade in ASCE’s (American Society of Civil Engineers) 2013 report card for America’s infrastructure was a D+, with a 3.6 trillion dollar estimated investment needed by 2020 and needs for inspection and assessment that far surpass available manpower. Crowdsourcing is increasingly being seen as one potentially powerful way of increasing the supply of labor for problem solving tasks, but there are a number of concerns over the quality of the data or analysis conducted. This is a significant concern when dealing with civil infrastructure for obvious reasons: flawed data could lead to loss of lives. Our goal was to determine if workers on Mechanical Turk were capable of developing basic engineering analysis skills using only the training afforded by comprehensive tutorials and guided questionnaires. Crowdsourcing has been effectively applied in the sciences, even prior to the Internet. The Audubon society has been harnessing the power of the crowds in order to effectively plot the location of hundreds of bird species in the United States. Thousands of Audubon members would mail information stating the number, species of birds, and their locations. Now the Audubon society and the Cornell lab of Ornithology run a real-time, online checklist program called eBird. Some other famous instances of effective citizen science include Galaxy Zoo, a galaxy classifying website and Phylo a game that allows crowds to help align related DNA sequences. Our goal was to test the possibility of Citizen Engineering for complex engineering tasks.

MEETING & EVENT ROOM SPECIFICATIONS // MAIN LEVEL MCKENNA HALL

MEN WOMEN

ELEVATOR

114 AUDITORIUM

WELCOME ATRIUM DESK 112

ELEVATOR

REGISTRATION

BUSINESS CENTER

108 106 104 102 100 109

HOLLOW ROOM SIZE THEATER CLASSROOM BOARDROOM U-SHAPE SQUARE 100 22' x 22' 25 16 16 12 16 102 32' x 22' 50 28 28 16 24 104 22' x 22' 25 16 16 12 16 100-102 54' x 22' 60 36 36 36 40 102-104 54' x 22' 60 36 36 36 40 100-104 22' x 76' 110 84 56 52 56 106 24' x 22' – – 16 – – 108 15' x 22' – – 12 – – 109 24' x 18' – – 6 – – 112 24' x 20' 35 24 16 10 12 114 24' x 20' 35 24 16 10 12 112-114 24' x 40' 70 48 32 24 28 Auditorium 70' x 50' 296 – 62 – –