Meeting 27: Statistical Learning

Reading: AIAMA 20.1-20.2

Recall from lecture 25 that linear classifiers could be derived from Bayes rule under specific assumptions about Gaussian distributions and equal covariances. This section describes how to relax that in order to learn parameters from arbitrary distributions, in effect to learn the parameters of a Bayesian Network. Since these techniques are the same as those used in statistics, these methods are called statistical learning methods.

Questions you should be able to answer after reading are:

1. What is Bayesian learning?
2. What is the difference between the posterior, likelihood, prior, and evidence terms?
3. How does maximum likelihood estimation (MLE) work?
4. How does maximum a-posteriori estimation (MAP) work?

Links

• Slides