|
I have the graphical model which is represented as some complex graph with sizes of maximum cliques more than 2. However, I want to define the energy of the model as sum of unary node energies plus sum of binary energies between connected nodes. In MRF the full energy must be defined as sum of max-clique energies. I'd like to understand if any inference methods for MRF could be applied in this case. Could someone give the starting point to think on how could I effectively minimize such energy? |
|
I would say -somebody correct me if i'm wrong- that in practice you dont have to define your MRF energy as a sum of max-cliques, because having pairwise interactions is a special case of having energies defined over max-cliques. So you can just solve a MRF where your cliques are is the set of pairwise connected nodes. |
|
As Alexandre says, using factor graphs and representing the cliques with that is the standard approach to do inference in MRF's There was a couple of good lectures on the Summer School on MRF and how to do inference on them. The slides are freely available at: |
|
Just write your model in terms of factor graphs. |
