Information Theoretic Sensitivity Analysis for Stochastic Simulators

• Activity Number: 248
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yu-Jay Huoh*+ and Cari Kaufman
• Companies: University of California, Berkeley and UC Berkeley
• Keywords: Bayesian ; computer models ; uncertainty quantification ; sensitivity analysis ; entropy

Abstract

The increased computational power available today has made the use of "computer models" or "simulators" common in many fields. While there is a widely adopted set of tools for the analysis of simulators, there are still many unsolved problems when dealing with these models. Specifically, the traditional methods for sensitivity analysis of computer models, based on functional ANOVA decompositions, do not generalize well to simulators with stochastic or nondeterministic output. This paper presents a methodological solution for conducting sensitivity analysis on computer models with stochastic output through the use of information theory and Bayesian density regression.