Statistics Seminar: STA 290
THURSDAY, May 22nd 2014 at 4:10pm, MSB 1147 (Colloquium Room)
Refreshments at 3:30pm in MSB 4110 (Statistics Lounge)
Speaker: Ana Kupresanin, Lawrence Livermore National Lab
Title: "Statistical inference for fractional diffusion processes"
Abstract: Science-based computational models are widely used to predict the behavior of complex physical systems across almost all areas of engineering, physical, environmental, social and health sciences. Advances in algorithms for solving mathematical systems and in computer speeds are making it possible to replace some physical experiments with computer experiments. In a computer experiment, a deterministic output is computed for each set of input variables. However, computational results almost always depend on inputs that are uncertain and on approximations that introduce errors, and are based on mathematical models that are imperfect representations of reality.
This presentation will give an overview of the fields of verification and validation (V&V) and uncertainty quantification (UQ), focusing on the types of experimental and computational data that are common to the LLNL's engineering applications. I will point out ways in which statisticians can collaborate with subject matter experts. For example, assuming that the computational model output can be modeled as a realization of a Gaussian Stochastic Process and designing computer experiments sequentially can lead to a better use of allocated computational resources. Given inevitable flaws and uncertainties in the results of a computational model, one's confidence in the results must originate from an understanding of the model’s limitations and uncertainties inherent in its predictions. Therefore I argue that a collaboration between subject matter experts and statistician is critically important and that they should jointly educate policy makers who make decisions based on computational models.