loopy BP as variational inference

(2.8 hours to learn)


Loopy belief propagation sounds like a hack, but it can be interpreted as a variational inference algorithm. In particular, it is a fixed point update for an approximation to variational inference, where both the energy functional and the marginal polytope are approximated. While this analysis doesn't lead to any strong guarantees, it is the basis for generalizations of loopy BP which have stronger guarantees.


This concept has the prerequisites:

Core resources (read/watch one of the following)


Graphical models, exponential families, and variational inference (2008)
An in-depth review of exact and approximate inference methods for graphical models.
Authors: Martin J. Wainwright,Michael I. Jordan

Supplemental resources (the following are optional, but you may find them useful)


See also