Experimental Design

Note

In-depth information about the theoretical underlying and the calculation methods will be described in a book chapter releasing in the near future.

The Experimental Design goal is to maximize the objective function based on one of the properties of the Fisher information matrix (FIM) \(F\), which is inversely proportional to the minimal squared estimation error [1]. Different optimality criteria (properties of the FIM) can be chosen. Some of the most popular criteria are:

1. D-optimality criterion

   Maximizes the determinant \(\det (F)\) of the FIM.

2. E-optimality criterion

   Maximizes the minimal eigenvalue \(\lambda_{\min}\).

3. A-optimality criterion

   Maximizes the sum of all eigenvalues \(\sum_i \lambda_i\).

4. Modified E-optimality

   Maximizes the ratio between the minimal and maximal eigenvalue \(\lambda_{\min} / \lambda_{\max}\).

[1]

Roy Frieden and Robert A Gatenby. Exploratory Data Analysis Using Fisher Information. Springer Science & Business Media, London, May 2010. ISBN 978-1-84628-777-0. doi:10.1007/978-1-84628-777-0.