TILS5110 Topics in computational statistics (2–15 cr)

Gaussian processes are a flexible tool for nonparametric nonlinear Bayesian regression, widely used in e.g. spatial statistics and probabilistic machine learning. This is a course on the practical and computational aspects of Gaussian process (GP) inference, with the main focus on the choice and implementation of GP models appropriate for the particular task at hand. The prerequisites in linear algebra and MCMC methods will be reviewed as needed. Time permitting, further topics can include composite models (i.e. hierarchical models with a GP component), various approximate inference methods for GPs and Bayesian optimization.