Speaker
Florian Kaspar
Description
Physics analyses often involve complex statistical models. When more than one model is available, it may be desirable to choose the "best" one of them. In a Bayesian setting, one decides by comparing model evidences.
For models with large numbers of parameters, calculating the evidence involves high-dimensional integrals.
Modern Monte Carlo methods can sample from high-dimensional probability density functions.
We apply restricted harmonic-mean integration with these samples as a way of calculating model evidences.
We present the method and where it is applicable, estimate its uncertainties and discuss pathological cases and possible extensions.
