On Bayesian approach to experimental design

Abstract

Design of experiment has been widely applied in the fields of science and industry but an excellent design of experiment diverges the ratio of extracted information to invested resources, so it is necessary to obtain an optimal design. The Bayesian approach plays an important role to solve this problem, as it treats model parameters as random variables rather than constants. An optimal design can be obtained by optimizing the predicted utility of the experiment. Two design models, A- and D-optimal designs are presented in this thesis. A-optimality minimized the sum of the main diagonal elements of the information matrix and D-optimality maximized the determinant of the information matrix.