0

In a graphical model with variables with continuous distriubtions, and some observed variables, how can I compute the messages to be passed? I know the messages but I don't know how to implement it? For continuous variables, and also for observed ones? Can anyone introduce a source which has the theory and an example implementation?

user3034939
  • 77

1 Answers1

1

Expectation propagation is what you are looking for. BP is a special case of EP where variables are discrete.

Mo Chen
  • 31
  • But what about loopy belief propagation? Is that discrete too? The formula of message passing in this page: http://research.microsoft.com/en-us/um/cambridge/projects/infernet/docs/How%20to%20add%20a%20new%20factor%20and%20message%20operators.aspx It is said that updates come from belief propagation: http://stats.stackexchange.com/questions/121180/compute-expectation-propagation-messages-for-sum So belief propagation seems to have a continuous representation. Also I don't understand this continuous case (EP or BP) for observed variables case. How we integrate observed variables in the model? – user3034939 Nov 17 '14 at 09:33
  • 1
    Here also we can see there exists variants of for example Gaussian Belief Propagation: http://en.wikipedia.org/wiki/Belief_propagation – user3034939 Nov 17 '14 at 09:38