A note on properties of using Fisher information gain for Bayesian design of experiments

03/16/2020

∙

by   Antony M. Overstall, et al.

∙

0

∙

share

Designs found by maximizing the expected Fisher information gain can result in a singular Fisher information matrix. This leads to non-unique classical estimates and ill-conditioning of posterior computation. A mitigating strategy for finding designs using Fisher information gain is proposed.