Skip to main content
U.S. flag

An official website of the United States government

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

ACMD Seminar: Uncertainty Quantification and Inversion of Large Models

Prof. Tony O'Hagan
Emeritus Professor of Statistics, School of Mathematics and Statistics, The University of Sheffield

Tuesday, October 10, 2017, 15:30 - 16:30
Building 101, Lecture Room A
Gaithersburg

Tuesday, October 10, 2017, 13:30 - 14:30
Room 1-4058
Boulder

Host: Dr. Jeffrey Fong

Abstract: Complex computational models are employed in almost all areas of science, engineering, industry and even government, in order to predict the behaviour of real-world systems.  They are increasingly used to inform commercial or policy decisions with high impacts.  And users are therefore increasingly asking how accurate the predictions from such models are.

All models are imperfect and there are many reasons why the predictions will differ from the true values in the real-world system.  Therefore, for a given prediction there is uncertainty about the true value, and models do not deserve to be trusted unless the uncertainty is quantified.  To do so, it is necessary to quantify all components of uncertainty in the prediction, to create an uncertainty budget.

The field of Uncertainty Quantification has grown rapidly in recent years, ostensibly to address this question.  However, most of the work in the field has concentrated on parametric uncertainty, i.e. uncertainty in predictions due to uncertainty about the model inputs, which is only one of several sources of uncertainty.  I will begin this talk by presenting a list of uncertainty components, and showing how the use of statistical emulators can address all of these.

The converse of prediction is inversion - observations in the real-world system are used to learn about model inputs.  Again, commonly used methods do not take account of all sources of uncertainty.  The second part of my talk will show how this leads to bias and over-confidence regarding the parameters of interest.

Bio: Tony O'Hagan obtained his BSc (1969) and PhD (1974) in Statistics from the University of London. Between these two bouts of study, he worked for two years as a practising statistician at the Central Electricity Generating Board. He began his academic career at the University of Dundee in 1973, moving to Warwick University in 1975. He became Professor of Statistics at the University of Nottingham in 1990 and moved to Sheffield in 1999. In 2008 he formally retired from his post at Sheffield, but stayed on part-time until September 2012 to complete the "Managing Uncertainty in Complex Models" project. He is an Emeritus Professor at The University of Sheffield.

Prof. O'Hagan has been able to serve the academic statistics community in a variety of capacities. In particular, he served on the Research Section Committee and the Council of the Royal Statistical Society, and the Board of the International Society for Bayesian Analysis. He was a member of the Peer Review College of the Engineering and Physical Sciences Research Council and the College of Experts of the Medical Research Council. He was a member of the editorial board of Significance Magazine for ten years from its beginning in 2004.

Prof. O'Hagan's research is in the theory and applications of Bayesian statistics. His main areas of research and consulting activity are in the elicitation of expert knowledge, in managing and quantifying uncertainty in the use of complex mechanistic models, and in Bayesian modelling generally. He has been involved in many application areas, particularly in medicine, environmental statistics, asset management and health economics.

Being retired from academic work frees Prof. O'Hagan to be more active as a consultant to government and industry. He is a Chartered Statistician, which is a qualification administered by the Royal Statistical Society that recognises his expertise as a practising statistician. He is a Fellow of the International Society for Bayesian Analysis and a Fellow of the Royal Statistical Society.

Contacts

Created September 25, 2017, Updated November 15, 2019