Bayes linear analysis of risks in sequential optimal design problems (Q1711561)

The main aim of the paper is to develop an approximation framework which provides decision support for the sequential design problem. A procedure is designed to be able to cope with problems that have large numbers of stages, as well as problems in which the system model is an expensive function that may only be evaluated at a handful of parameter settings. The proposed procedure is designed to track uncertainty which is a consequence of approximation, throughout the calculation. It allows the user to make an informed decision about whether to select an experiment subject to these uncertainties, or to carry out further analysis which may reveal more about the risks involved. A similar framework to handle non-sequential design problems for expensive models was proposed by \textit{M. Jones} et al. [J. Stat. Plann. Inference 171, 115--129 (2016; Zbl 1336.62209)].