ENBIS-16: Introduction to Kriging using R and JMP

11 September 2016; 09:00 – 16:00; University of Sheffield, School of Mathematics & Statistics, Hicks Building (Room TBA)

ENBIS-16 Full-Day Pre-Conference Workshop on Sunday, September 11th, 2016

 

Workshop Description

Kriging (or Gaussian process regression) has proven to be of great interest when trying to approximate a costly to evaluate function in a closed form. The main aim of the workshop is to show how to build useful surrogate models using this approach, and to make clear the assumptions that such models rely on. Furthermore, once it exists, we will show how a surrogate model can be used for optimization. In the morning we introduce the concepts of kriging through lectures and demonstrations. During the afternoon we apply some of the key methods in lab sessions using both R and JMP. The main aim of the lab is to quickly find optimal settings of a catapult numerical simulator that can fire the longest shot.

The workshop is designed for people with applications that require them to work with functions, often of many variables, that are costly to evaluate. Knowledge of linear regression, statistical modeling, and stochastic processes is helpful but is not required for the workshop. Similarly, basic knowledge of R and/or JMP will be helpful for the hands-on lab component, but is not mandatory.

The workshop should enable participants to:

  • understand how to build and validate Kriging models;  
  • understand the assumptions behind Kriging models;  
  • appreciate the influence of the parameters in Kriging models  
  • gain experience of designing and analyzing computer experiments in R and JMP.

 

Workshop Outline

Morning (9-12):

  • An Introduction to Kriging.
  • Surrogate models (or emulators) in engineering.
  • Kriging (Gaussian process regression). 
  • Covariance functions.
  • Parameter estimation.
  • Model validation.
  • Application to optimization.
  • Design of experiments.

Afternoon (13-16)

  • Lab Session.
  • Optimization of a 4-dimensional function (a catapult numerical simulator) with R and JMP.
  • Optimization of a higher-dimensional problem - JMP.

 

Teaching Materials Provided to Course Participants

  • Lecture slides
  • R codes
  • JMP journal, data and scripts

 

Technical Prerequisites for Participants

Participants should bring their own laptops on which R has to be pre-installed, ideally with the following packages: DiceOptim, DiceView, DiceDesign and Shiny. JMP 12 (or JMP Pro 12) should also be pre-installed (JMP-R integration requires Windows). A free 30-day trial version is available at www.jmp.com/trial.

 

Workshop Facilitators

Nicolas Durrande is a lecturer at Mines St-Etienne (France). His research interests include Gaussian process models and kernel theory. He is co-organizer of the OQUAIDO project (Chair of applied mathematics at Mines St-Etienne), which comprises thirteen industrial and academic partners working jointly on the topic of computer experiments.

Volker Kraft is the Academic Ambassador for JMP in Europe, fostering and supporting the use of JMP in teaching and research. He holds a Ph.D. in Electrical Engineering (University Bochum, Germany), and used statistical methods extensively in his research in psychoacoustics and speech communication.