WSC 2004 Final Abstracts
Monday 10:30:00 AM 12:00:00 PM
Development Support for Agents
Chair: John Hiles (MOVES Institute)
Peeking Into the Black Box: Some Art and Science to Visualizing Agent-Based Models
Stephen M. Guerin (RedfishGroup)
This paper explores current metaphors for visualizing agent-based models. Metaphors include grid, network, n-dimensional cubes and landscape visualization techniques. A final section offers some theory underlying visualization of complex systems models with emphasis on mappings to non-equilibrium systems; conserved quantities and their flows; identifying order parameters and control parameters; and the presentation of phase transitions.
Vigilance Performance Modeled as a Complex Adaptive System with Listener Event Graph Objects (LEGOS)
Joerg C.G Wellbrink (Federal Office of the Bundeswehr for Information Management and Information Technology) and Arnold H. Buss (The MOVES Institute)
There has been an increasing need to incorporate human performance in simulation models. Situations in which human performance is subject to degradation over time, such as vigilance tasks, are not represented. This article describes a computational model for vigilance performance embedded in a new cognitive framework that utilizes recent advances in system neuroscience, evolutionary psychology and complexity theory. The Reduced Human Per-formance Model (RHPM) captures human errors in monitoring tasks to a greater degree than previous at-tempts. RHPM is implemented as a discrete event simulation using Listener Event Graph Objects (LEGOs). The model captures leading vigilance theories and can be used as a tool to improve existing vigilance theories and to improve current monitoring procedures minimizing errors that could lead to catastrophic outcomes.
IAGO Project and Development of Compound Agents
John E. Hiles (MOVES Institute)
The IAGO Project explores the question of whether a software model, in the form of a computational model of cognitive behavior, can contribute to better anticipation of asymmetrical threats. The computational model used in IAGO is based on Cognitive Blending, a theoretical model proposed in the Cognitive Sciences to explain fundamental or backstage cognitive operations in the brain. This model was implemented with the use of multiagent systems that coordinated their activity with a bio-inspired operator called a Connector. This operator and several others used in the IAGO project have been incorporated into a pro- gramming library, called the CMAS Library. CMAS stands for Compound Multiagent System. Compound re- fers to multiagent systems, in which at least some of the agents contain embedded multiagent systems. In the case of IAGO these embedded systems implement Cognitive Blending.
Monday 1:30:00 PM 3:00:00 PM
Agent Modeling Techniques I
Chair: Andy Hernandez (Naval Postgraduate School)
Recently there has been considerable interest in Multi-agent systems(MAS). However, it is very difficult to accu-rately design and implement simulation of MAS, in respect that the systems are often extremely complex and based on autonomous software and hardware components, termed as agents, which cooperate within an environment to perform some task. Obviously interoperability is one of the key roles to play in the investigation and development of Multi-agent systems. This paper begins with introduction to High Level Architecture; following, provides an emphatic intro-duction to agent architecture. Then we outline an approach how to make the MAS simulation accord with the HLA framework. Especially show how the agents simulated on different machines can become a legal Agent Federate joining the flexibly MAS simulation federation. Also em-phatically describe the simulation cycle of the MAS system.
Exploring Agent-Supported Simulation Brokering on the Semantic Web: Foundations for a Dynamic Composability Approach
Levent Yilmaz (Auburn University) and Tuncer I. Íren (University of Ottawa)
Federated simulations address the need for interoperability, as well as the improvement of reuse and composability. The focal goal in a federated simulation is to facilitate composable simulations by standardizing interfaces to assure technical interoperability among disparate simulations. Yet, existing federated simulation infrastructures neither facilitate substantive interoperability nor are dynamically extensible. Emergent web services technologies hold out the potential to significantly improve the development of interoperable, extensible, and dynamically composable federations. As such, recent initiatives (i.e., XMSF) are urging the use of open standards that can be applied within an extensible framework for next generation modeling and simulation applications. We discuss how the realization of multimodel and multisimulation formalisms in terms of semantic web and agent technologies may bring new vistas to demonstrate runtime model discovery, instantiation, composition, and interoperation.
Simulating Growth Dynamics in Complex Adaptive Supply Networks
Surya Dev Pathak, David M Dilts, and Gautam Biswas (Vanderbilt University)
This paper discusses an extended adaptive supply network simulation model that explicitly captures growth (in terms of change in size over time, and birth and death) based on Utterback’s (1994) industrial growth model. The paper discusses the detailed behavioral modeling of the key components in the model with the help of statechart and decision tree representations. The design of a distributed, multi-paradigm, agent-based simulation that addresses the issue of scalability and computational efficiency is pre-sented. The system is targeted to run on a supercomputing grid infrastructure at Vanderbilt University. We present a method for validating this model using an experimental design that models the growth dynamics of the US auto-mobile industry supply network over the past 80 years. The experimental work is now in progress and the results and analysis of this work will be presented during the con-ference.
The Parallel and Distributed Simulation of Mas Base on HLA Framework
Wang Xue hui (School of Mechatronics Engineering and Automation) and Zhang Lei (Department of Computer Science)
Monday 3:30:00 PM 5:00:00 PM
Agent Modeling Techniques II
Chair: Aaron Van Alstine (Naval Postgraduate School)
Resolving Mutually Exclusive Interactions in Agent Based Distributed Simulations
Lihua Wang, Stephen John Turner, and Fang Wang (Nanyang Technological University)
With the properties of autonomy, social ability, reactivity and pro-activeness, agents can be used to represent entities in distributed simulations, where fast and accurate decision making is a determining factor of the whole environment. Resolving concurrent interactions is a key problem of this kind of system, as the shared environment needs to allow agents to interact with the environment in a causally consistent way. There will usually be either mutually exclusive or collaborative interactions. This paper presents our research in designing a middleware component called Interaction Resolver (IR) to resolve the effect of concurrent interactions and still guarantee the consistency and causality of the system. The ownership management services provided by the High Level Architecture (HLA) are compared with IRs in resolving mutually exclusive interactions in our prototype, a minesweeping game. Conclusions are drawn based on the experimental results.
Data Dissemination Techniques for Distributed Simulation Environments
Bryan Horling and Victor Lesser (University of Massachusetts)
Farm is a distributed simulation environment for modeling the performance of large-scale multi-agent systems. It uses a component-based architecture to distribute the computational load of the simulation and improve running time. It also supports a global data repository, which permits both actors running in the simulator and external analysis components to generate and use arbitrary pieces of information. Because the components are distributed, the manner in which this data is accessed can have significant effect on the communication overhead and duration of the simulation. In this paper we explore several different techniques for accessing and disseminating this data. Analytic and empirical models of the system's performance are presented, along with an analysis of which strategy is appropriate under different conditions.
Behavioral Anticipation in Agent Simulation
Tuncer I. Íren (University of Ottawa) and Levent Yilmaz (Auburn University)
In this article, the following is done: (1) a systematic and comprehensive classification of input is given and the relevance of perception as an important type of input in intelligent systems is pointed out, (2) a categorization of perception is given and anticipation is presented as a type of perception, (3) the inclusion of anticipation in simulation studies is clarified and other aspects of perceptions in simulation studies especially in conflict situations are elaborated.
Tuesday 8:30:00 AM 10:00:00 AM
Chair: Gary Horne (The MITRE Corporation)
Data Farming: Discovering Surprise
Gary E. Horne and Ted E. Meyer (The MITRE Corporation)
The development of models and analysis of modeling results usually requires that models be run many times. Very few modelers are satisfied with the computing resources available to do sensitivity studies, validation and verifica-tion, measurement of effectiveness analysis, and related necessary activities. Fortunately, high performance com-puting, in the form of distributed computing capabilities and commodity node systems, is becoming more pervasive and cost effective. In this paper the authors describe the concept and methods of Data Farming, the study and de-velopment of methods, interfaces, and tools that make high performance computing readily available to modelers and allows analysts to explore the vast amount of data that re-sults from exercising models.
Simulation in Context: Using Data Farming for Decision Support
Philip Barry and Matthew Koehler (The MITRE Corporation)
Data Farming leverages high performance computing to run simple models many times. This process allows for the exploration of massive parameter spaces relatively quickly. This paper explores a methodology to use Data Farming as a decision support tool. Data Farming can be a highly ef-fective in this role because it allows one to present to a de-cision-maker not only what may be the most likely out-come but what are possible outcomes, especially outliers that might have far reaching impact. The terrorist attacks of September 2001 are a good example of an outlier with very high impact. A case study is presented using a simple terrorist attack simulation and decision-maker utility model.
Data Farming Coevolutionary Dynamics in RePast
Brian F. Tivnan (The MITRE Corporation)
This paper describes the application of data farming techniques (Brandstein and Horne 1998) to explore various aspects of coevolutionary dynamics (McKelvey 2002) in organization science. Data farming is an iterative process using high-performance computing to execute and vary agent-based models, collect and explore statistical results, and integrate these results for the purposes of growing more data by virtue of generative analysis. The tool of choice for creating these agent-based models is the University of Chicago's Social Science Research Computing’s (2004) REcursive Porous Agent Simulation Toolkit (RePast). The paper concludes with a brief description of Tivnan’s (2004) Coevolutionary model of Boundary-spanning Agents and Strategic Networks (C-BASN), an extension of Hazy and Tivnan’s (2004) Model of Organization, Structural Emergence, and Sustainability (MOSES).