Speaker
Jan Hamann
(The University of New South Wales)
Description
How can we decide which cosmological model is the most probable? In a Bayesian approach to statistics, this question can be readily answered using the framework of Bayesian model selection: namely by calculating a model's evidence. However, the numerical evaluation of the evidence can be a numerically difficult task.
I will introduce a new, efficient machine-learning approach to this problem – based on Gaussian Process Regression and Bayesian Optimisation and designed to minimise the number of likelihood evaluations required – and demonstrate its efficiency on a number of examples.
Track type | Cosmology |
---|
Author
Jan Hamann
(The University of New South Wales)