Graphical Models and Inference - Lectures 15 and 16

Bayesian Graphical Models

This lecture describes the general idea behind modern developments in Bayesian statistics, using graphical models as a paradigm in combination with Markov chain Monte Carlo Methods (MCMC) for computation.

The idea is to treat ordinary statistical problems as special instances of probabilistic expert systems, replacing probability propagation algorithms with ordinary MCMC. Parameters and data are explicitly represented in the network and plates are used to indicate repetition patterns.

Bayesian graphical models are very well described in the manual to WinBUGS.

Here is a link to a draft chapter in a forthcoming book about Graphical Models in R

Overheads

Links

Previous lecture

Course overview


Last updated: Monday, 28 November 2011Steffen L. Lauritzen