Experiment Design
for Models with Field and Computer Trials
Derek Bingham,
Simon Fraser University
Pritam Ranjan,
Simon Fraser University
Shane Reese, BYU
Physical
experimentation is an expensive endeavor. Computer experiments,
which are often a
less expensive alternative, can provide a reasonably
accurate
representation of physical experimental. This work proposes an
approach to
designing experiments when the researcher has a limited amount of
data collected on
each of the physical and computer systems and wishes to
allocate future
experiments for both the physical and computer systems. This
hybrid
experimental situation relies on a Gaussian process formulation for
the modelling of
experimental outputs as functions of the system parameters
inputs. We
demonstrate the methodology with both simulated and actual
experimental
situations.