1: \begin{abstract}

2: We elaborate on the idea that loop corrections to belief propagation 

3: could be dealt with in a systematic way on 

4: pairwise Markov random fields,

5: by using the elements of a cycle basis to define region in a generalized belief propagation setting.

6: The region graph is specified in such a way as to avoid dual loops as much as

7: possible, by discarding redundant Lagrange multipliers, 

8: in order to facilitate the convergence, while avoiding instabilities 

9: associated to minimal factor graph construction.

10: We end up with a two-level algorithm, where a belief propagation algorithm is run 

11: alternatively at the level of each cycle and at the inter-region level.

12: The inverse problem of finding the couplings of a Markov random field from empirical covariances

13: can be addressed region wise. It turns out 

14: that this can be done efficiently 

15: in particular in the Ising context, where fixed point equations can be derived along with 

16: a one-parameter log likelihood function to minimize. Numerical experiments confirm the effectiveness of

17: these considerations both for the direct and inverse MRF inference. 

18: \end{abstract}

19: