Graphical Models
Table of contents
GMs are graph based representations of various factorisation assumptions of distributions. These factorisations are typically equivalent to independence statements amongst (sets of) variables in the distribution.
Bayesian Networks
Each factor is a conditional distribution. Generative models, AI, statistics. Corresponds to a DAG.
Markov Networks
Each factor corresponds to a potential (non negative function). Related to teh strength of relationship between variables, but not directly related to dependence. Useful for collective phenomena such as image processing. Corresponds to an undirected graph.
Chain Graph
A marriage of BNs and MNs. Contains both directed and undirected edges.
Factor Graph
A barebones representation of the factorisation of a distribution. Often used for efficient computation and deriving message passing algorithms.