Abstract 

Simulation codes are used to analyze systems where the complexity of physical experimentation prohibits construction of usable inference from raw data.  The challenge in using these codes is two-fold: (i) the models do not match exactly with the true physical system and (ii) the codes require a nontrivial computational cost thus we have a limited number of evaluations of the code.  Dealing with these aspects simultaneously is the major challenge of the field of computer experiments, and from a larger perspective, uncertainty quantification.

The first segment of the talk will discuss a method to study the effect of mutations on the action potential, the fast electrical activity after an applied membrane potential.  In this case, our parameter of interest is a function as opposed to a scalar or a set of scalars.  We develop a new modeling strategy to nonparametrically study the functional parameter using Bayesian inference with Gaussian process priors.  Methodological challenges mainly center on Bayesian computation, which were addressed through a mix of existing and new algorithms for Markov chain Monte Carlo analysis.

The second part of this talk will discuss new computer experiments that yield fast-to-compute predictors. Statistical frameworks avoid repeated evaluations of the code through an emulator, constructed by conducting an experiment on the code. In high dimensional scenarios, the traditional framework for emulator-based analysis can fail due to the computational burden of inference.  In this talk, we will propose a new class of experiments, termed space grid designs, that offer important results in two areas: (i) inference from half a million observations is possible in seconds versus the days required for the traditional technique and (ii) the designs achieve predictive accuracy on par with traditional space filing designs.

Bio 

Matthew Plumlee is an assistant professor in the Department of Industrial and Operations Engineering at the University of Michigan. Plumlee’s research interests focus on analytics for product and process improvement.  During his time as a PhD student at Georgia Tech, where he also received an MS in Statistics,  Plumlee was awarded the INFORMS Quality, Reliability and Statistics best student paper award, the Natrella Scholarship and the Ellis R. Ott Scholarship.  Matthew’s research interests are in statistical methods, with a focus on with a focus on uncertainty quantification in modeling, analysis, and decision making.  Portions of his dissertation work appeared in the Journal of the American Statistical Association and Technometrics.