Description
Cosmological model selection, in the framework of Bayesian inference
requires the calculation of the Bayesian evidence. This can be challenging if the
underlying likelihood function is slow to evaluate, which often happens when
accurate modelling of cosmological observables requires complex and
time-consuming numerical calculations. I will discuss how a technique called
Bayesian Optimisation, based on Gaussian Process regression, can be used to
calculate this evidence as well as parameter posteriors in far fewer likelihood
evaluations, offering a much more efficient approach compared to traditional
methods for such likelihoods. Thus, inference tasks which would take days with
traditional methods can be done in a few hours using this approach, opening up
the possibility to search for new physics across a wider range of cosmological
models and datasets.
