|
Abstract:
|
Many military offices (e.g., DMSO) rely upon complex simulations for planning. These simulations may involve agent-based models, systems of differential equations, or complex hybrids. Also, these simulations typically involve large numbers of parameters and are highly nonlinear. This means traditional use may be unreliable; even sensitivity analysis is hard to do. This talk will consider methods for understanding and assessing such models based on the intrinsic parameterization, local low-dimensional behavior, and parsimony. No method is perfect in such situations, but there are strategies for testing face-validity and obtaining partial corroboration of results.
|
