Understanding message passing algorithm and related theory

FRosner / MasterThesis

0 stars 0 forks source link

Understanding message passing algorithm and related theory #1

Closed FRosner closed 10 years ago

FRosner commented 10 years ago

Topics:

directed graphical models
undirected graphical models
factor graphs
variable elimination
message passing
variational inference / mean field

Sources:

Youtube, Murphy, Bishop

FRosner commented 10 years ago

Exact inference for graphical models:

Belief propagation for trees
The variable elimination algorithm
The junction tree algorithm

FRosner commented 10 years ago

I got what belief propagation / sum-product algorithm and the variable elimination algorithm is doing. I still have to look at the junction tree algorithm.

One open question is how the factors and the probabilities are related in general and what factor graphs have to do with graphical models. I should look at section 8.4 in Bishop (pg. 393) "Inference in Graphical Models".

FRosner commented 10 years ago

Variational message passing is implemented in the open source program VIBES (http://vibes.sourceforge.net). But I probably don't need variational message passing right now since the algorithm I need to implement is the sum-product algorithm.

FRosner commented 10 years ago

I think I got the main idea of the sum-product algorithm. I did not look deeper into the junction tree algorithm.