Docent lecture: "Choosing the next experiment: an introduction to Bayesian experimental design"

The more time and resources an experiment requires, the more important the choice of which experiment to perform. Large gains in efficiency can be made by basing each new decision on the information gathered so far, that is, by treating the choice as a sequential decision problem.

In this lecture, I will show how Bayesian experimental design offers a natural way both to update a probabilistic model as new observations arrive and to make optimal decisions based on it. I will also show why the best choice depends on whether the goal is to find the best candidate or to map out the whole space.

Finally, I will show how the same ideas are used in self-driving laboratories in materials science.

The lecture is an obligatory teaching test for those applying for admittance as docent and should be accessible to students and others with undergraduate-level knowledge in the relevant field. The lecture lasts 40–45 minutes and is followed by a discussion. The lecture will be given in Swedish.