Computational Spectrum of Agent Model Simulation
- ORNL
The study of human social behavioral systems is finding renewed interest in military, homeland security and other applications. Simulation is the most generally applied approach to studying complex scenarios in such systems. Here, we outline some of the important considerations that underlie the computational aspects of simulation-based study of human social systems. The fundamental imprecision underlying questions and answers in social science makes it necessary to carefully distinguish among different simulation problem classes and to identify the most pertinent set of computational dimensions associated with those classes. We identify a few such classes and present their computational implications. The focus is then shifted to the most challenging combinations in the computational spectrum, namely, large-scale entity counts at moderate to high levels of fidelity. Recent developments in furthering the state-of-the-art in these challenging cases are outlined. A case study of large-scale agent simulation is provided in simulating large numbers (millions) of social entities at real-time speeds on inexpensive hardware. Recent computational results are identified that highlight the potential of modern high-end computing platforms to push the envelope with respect to speed, scale and fidelity of social system simulations. Finally, the problem of shielding the modeler or domain expert from the complex computational aspects is discussed and a few potential solution approaches are identified.
- Research Organization:
- Oak Ridge National Laboratory (ORNL); Center for Computational Sciences
- Sponsoring Organization:
- ORNL Program Development; ORNL work for others
- DOE Contract Number:
- AC05-00OR22725
- OSTI ID:
- 982107
- Country of Publication:
- United States
- Language:
- English
Similar Records
Scalable Entity-Based Modeling of Population-Based Systems, Final LDRD Report
Multi-Channel Large Network Simulation Including Adversarial Activity