Computing That Serves

The Power of Bayes: An Old Idea Revisited


Thursday, October 11, 2007 - 11:00am


Shane Reese, Associate Professor, BYU Statistics Department

Bayesian methods represent an important subset of statistical methods which allow for a broadening definition of "data". Diverse information such as high dimensional computer models, traditional experimental data, and expert opinion can all be coherently combined using probability models. Advances in computational methods for Bayesian models allow for a great deal of flexibility in the choice of probability model. In this talk I will present a brief tour of Bayesian methods with illustrations from two areas of my work. The first application will be the incorporation of computer experiments into a sophisticated prior model. The second application will be a model for supercomputer reliability. Each of these illustrates diverse information being integrated into a single, unified analysis. Further, we demonstrate the great predictive ability of an important subclass of Bayesian models called hierarchical models throughthe two case studies.