SC1: Adaptive Markov chain Monte Carlo
Author: Krzysztof Latuszynski (Warwick University)
Markov chain Monte Carlo algorithms (MCMC) provide a generic way of sampling from complicated probability distributions, such as posteriors of complex statistical models in Bayesian inference. However performance of generic MCMC, such as Metropolis or Gibbs samplers, and consequently reliability of posterior estimation, depends on the tuning parameters of the algorithms, and optimising them is time consuming, requires expert knowledge, and may be infeasible for high dimensional problems.
Adaptive MCMC addresses this issue and aims to optimise the algorithm tuning parameters on the fly, as simulation progresses and more properties of the sampling problem are becoming available.
I will discuss design and theoretical properties of Adaptive MCMC algorithms and illustrate their applicability with examples from Bayesian inference.
SC2: Bayesian computing with R-INLA
Author: Håvard Rue & Daniela Castro (KAUST)
In this short course, we will give an introduction to the R-INLA package which facilitates the fitting of a large range of complex statistical models by dramatically reducing computation time compared to MCMC. We will discuss the underlying ides behind these approximations, and then show how the package can be used in various applications, including some new additions for fitting extreme values.