B1640
Title: Bayesian optimal designs for misspecified models
Authors: Antony Overstall - University of Southampton (United Kingdom) [presenting]
Abstract: The optimal design of experiments is considered for the case of a misspecified linear model. Suppose post-experiment, Bayesian inference will be used under a parametric likelihood. Then, the inference will target the parameter values that minimise the Kullback-Leibler divergence between the model and the true data-generating process. These target parameter values depend on the design used: an unattractive property. The talk will discuss Bayesian design of experiment approaches to ensure that the target parameter values are close to desirable target parameter values, i.e. values that have a fixed physical interpretation.
