CI-P: Workshop on Enhancing a Large-scale Multiagent Simulation Tool

George Mason University, Fairfax VA

Investigators

Abstract

MASON is an open source massive multiagent simulation toolkit, that is, a set of software tools designed to make it easy to build simulations of large numbers of agents -- robots, people, organizations, animals, etc. -- interacting with one another in a complex fashion. Such toolkits are used by scientists to model everything from schools of fish, to swarms of flying robots, to the rise of historic trade routes in Asia. These kinds of models are becoming increasing popular both in the scientific and engineering communities as computer power is enabling the study of large numbers of "agents" to interact with one another in interesting and nontrivial ways. MASON is considered to lie at the "high performance" end of software of this type: it runs on everything from laptops to supercomputers. As such it has become a popular tool for simulating thousands, even potentially millions of agents interacting with one another. MASON is now at a turning point in its development. Models are being developed which are pushing it to its limits, MASON's complexity is proving daunting to newcomers, and it is increasingly attractive to use MASON in the context of massive model optimization. A MASON workshop is intended to solicit community and expert feedback and input on extending and improving MASON in at least three potential ways: 1. A massively distributed version of MASON enables the library to harness large clusters of computers for even bigger and more costly models. 2. Connecting MASON with Integrated Development Environments, and developing a lightweight model development language can make it much easier for non-experts to use MASON. 3. Integration with large-scale stochastic optimization toolkits enables scientists to use MASON as a kind of modeler's assistant: the modeler specifies the high-level rules he believes to be true or wishes to test, and provides an assessment procedure for model output, and then the optimizer proceeds to "fill in the gaps", hunting for the remaining model rules which produce models which optimize this assessment procedure. Similar tools enable sweeps of model parameters. Output from the workshop will go towards development of a community-based plan and methodology to improve MASON in these and other aspects. Tools and techniques like MASON are having an increasing and unusually broad impact on society at large: they're used in large economic and financial models, social models of terrorist networks, game development and special effects, air traffic control, and ecological models ranging from coral reef damage to the impacts of climate change.