In the previous chapter, we saw how we can represent a joint probability distribution (JPD) using a directed graph and a set of conditional probability distributions (CPDs). However, it's not always possible to capture the independencies of a distribution using a Bayesian model. In this chapter, we will introduce undirected models, also known as Markov networks. We generally use Markov networks when we can't naturally define directionality in the interaction between random variables.
In this chapter, we will cover:
The basics of factors and their operations
The Markov model and Gibbs distribution
The factor graph
Independencies in the Markov model
Conversion of the Bayesian model to the Markov model and vice versa
Chordal graphs and triangulation heuristics