文档介绍：Graphical Models Page 1 of 19
A Brief Introduction to Graphical Models and Bayesian
Networks
For a non-technical introduction to works, read this LA times article (10/28/96). For some of the technical
details, see my tutorial below, or one of the other tutorials available here. For the really gory details, see the AUAI homepage
(Association for Uncertainty in Artificial Intelligence), or my list of mended reading at the end. The UAI mailing list is
now archived and provides up-to-date information. I also maintain a list of software packages, including my own
Toolbox for Matlab.
Note that, despite the name, works do not necessarily imply mitment to Bayesian methods. Rather, they
are so called because they use Bayes' rule for probailistic inference (see below). For non-technical introductions to Bayesian
methods, see this Economist article (9/30/00) or this New York Times article (4/28/01). For more technical details, see Tom
Minka's excellent tutorial notes.
works are also closely related to influence diagrams, which can be used to make optimal decisions. The most
famous (non)-example is the Microsoft Window's paperclip. (The reason this is a non-example is that the shipped version, as
opposed to the research version, did not in fact use Bayesian methods.) For a more recent example, Microsoft's Mobile
Manager, see this Economist article (3/22/01).
"Graphical models are a marriage between probability theory and graph theory. They provide a natural tool for dealing with
two problems that occur throughout applied mathematics and engineering -- uncertainty plexity -- and in particular
they are playing an increasingly important role in the design and analysis of machine learning algorithms. Fundamental to
the idea of a graphical model is the notion of modularity -- plex system is built bining simpler parts. Probability
theory provides the glue whereby the parts bined, ensuring that the system as a whole is consistent, and providing