Practical implementation of OED
By Maarten Speekenbrink (University College London)
Abstract: This presentation will provide a tutorial on the implementation of Optimal Experimental Design (OED) in practice. Throughout, we will focus on a Bayesian OED framework where the objective of an experiment is to gain as much information about parameters of interest. These parameters may be those of a single model that we wish to estimate as precisely as possible (a model estimation problem) or the index of the "true" model in a set of competing models (a model selection problem). I will start with an introduction to Sequential Monte Carlo (SMC) as a general technique to estimate statistical models in cognitive science. I will then show how to use SMC estimates to optimize the design of an experiment in order to maximise the information it provides about the relevant parameters of interest. All code for implementing the examples will be provided in the R language for statistical computing.