SFU Canada Research Chairs Seminar Series: "Efficient Emulators of Computer Simulators of Photometric Red-shifts Using Compactly Supported Correlation Functions "
11:30 - 12:30
Rm10900
Dr. Derek Bingham
Department of Statistics and Actuarial Sciences, Simon Fraser University
Abstract
Computer simulators are often used to study real-world processes that are too difficult to observe directly. Experimenter's are often interested in building a statistical surrogate (an emulator) for the computer model to avoid constantly running the model at different input settings. Building an emulator for a computer simulator using standard Gaussian process models can be computationally infeasible when the number of evaluated input values is large. As an alternative, we propose using compactly supported correlation functions, which produce sparse correlation matrices that can be more easily manipulated. Following the usual approach of taking the correlation to be a product of correlations in each input dimension, we show how to impose restrictions on the correlation range for each input, giving sparsity, while also allowing the ranges to trade-o? against one another, thereby giving good predictive performance when the data are anisotropic. As an illustration, the method is used to construct an emulator of photometric red-shifts of cosmological objects. This is joint work with Salman Habib, Katrin Heitman (Los Alamos National Lab) and Cari Kaufman (UC Berkeley).
About the Speaker
Derek Bingham received his PhD from the Department of Mathematics and Statistics at Simon Fraser University in 1999, winning the Governor General's Gold. After graduation he joined the Department of Statistics at the University of Michigan as an Assistant Professor. Derek has been at Simon Fraser since the Fall of 2003 when he arrived as the Canada Research Chair in Industrial Statistics in the Department of Statistics and Actuarial Science. Since 2002, he has held a Faculty Affiliate position at the Los Alamos National Laboratory. The main focus of Derek Bingham's research is the development of statistical methodology for the design and analysis of industrial experiments. This work focuses on developing new methodology for: (1) Design and analysis of computer experiments; (2) Bayesian design and analysis of experiments in industrial problems such as optimal screening designs, response surface optimization and optimal robust parameter designs for product variation reduction; and (3) Design and analysis of fractional factorial plans for multi-stage processes. Most of the problems that motivate this research are the result of scientific collaborations. For example, recent work relating to experiments on computer simulators is the direct result of interaction with scientists at Los Alamos National Laboratory, the National Center for Atmospheric Research (Boulder, Colorado) and the Center for Radiative Shock Physics at the University of Michigan. Derek also is the Principle Investigator on the NPCDS funded project on the Design and Analysis of Computer Experiments for Complex Systems as well as NSERC Discovery and Accelerator grants on the design of industrial experiments. To facilitate interaction with industry, he helped create the CFI funded Industrial Statistics Laboratory within his Department. The Lab provides an environment for faculty, research scientists, students, post-docs and visitors to collaborate on research problems. SFU faculty interested in research collaborations are always encouraged to stop by. The Department of Statistics and Actuarial Science has several areas of strength, ranging from biostatistics and biometrics to survey sampling. In addition to developing fundamental theory and methodology, we are actively involved in applied problems that arise in such diverse fields as astronomy, biology, chemical engineering, finance, fisheries, forestry, medicine and space weather - only to name a few. We have forged strong interdisciplinary links with other departments and actively seek to recruit graduate students and faculty who can help to build and maintain such links.
