 |
 |
*_Notice and Invitation_* Oral Defense of Doctoral Dissertation The Volgenau School of Engineering, George Mason University Kenneth W. Comer Bachelor of Arts, Cornell University, 1974 Master of Arts, Georgetown University, 1983 Master of Science, The George Washington University, 1989 *Who Goes First? An Examination of the Impact of Activation on Outcome Behavior in Applied Agent-based Models* Monday, March 10, 2014, 10:00AM - Noon Room 2901 Nguyen Engineering Building All are invited to attend. *_Committee_* Dr. Andrew G. Loerch, Chair Dr. Chun-Hung Chen Dr. Rajesh Ganesan Dr. Robert Axtell _Abstract_ Agent-based models have become the tool of choice for modeling self-organizing systems. In fact, for some domains they have supplanted traditional discrete event simulations as decision-support elements to explore the potential outcome space. In creating a specific agent-based model, there are several choices the simulation designer must make. Often these design decisions are implicit, but they may be important in the performance of the simulation. One such choice is the sequence with which the agents will execute their methods or change their state. This is the 'activation' question, and its impact on three different models is examined in this dissertation. Three agent-based models described in the literature in three separate domains (civil unrest, market execution, and social interactions) were replicated, and the impact of various activation schemes on the emergent population patterns and dynamics was analyzed. It was demonstrated that the choice of activation type is important for the outcome behavior of the model and should be stipulated in any published description of an agent-based model. In some experiments the differences noted, while significant, were only statistical. In others they led to substantial differences in either outcomes or model behavior. Further investigation showed that sophisticated activation schemes can become powerful tools to produce unexpected or unpredicted behavior of multi-agent systems. Thus, activation becomes more than an inconvenient detail to be dealt with during design, and is shown to be a source of exploratory variation as modelers of self-organizing social systems seek to match the behavior of natural systems. A copy of this doctoral dissertation is on reserve at the Johnson Center Library.