Graphical Models and Inference - Lecture 4

Maximum likelihood in log-linear models

This lecture discusses the problem of finding the MLE in log-linear models for contingency tables. We show uniqueness and existence (in the closure of the model) , show the MLE is the solution of a system of equations which equates the observed and theoretical marginals to generators, and describe the fundamental algorithm of Iterative Proportional Scaling. For a more detailed description of this you way wish to consult my ancient Lectures on Contingency Tables.

Overheads

Links

Next lecture

Previous lecture

Course overview


Last updated: Monday, 24 October 2011Steffen L. Lauritzen