APPROXIMATE INFERENCE

摘要

Method for computing at least one marginal probability for an observed phenomenon in a probability model which describes probabilities for relationships between said phenomenon and physical parameters by using a Cluster Variation Method (CVM), said model comprising a set of variables xi said variables xi representing observations of said physical parameters, the set of variables including a plurality of subsets; the marginal probability including a subset of the set of variables, the probability model being defined by a probability distribution ppsi(x), the probability distribution ppsi(x) having a potential psi(xi) for each of the subsets; the method (CVM) including: - defining a first cluster alpha of a first cluster-type, at least one second cluster beta of a second sub-cluster type and at least one third cluster gamma of a third sub-cluster type, each being one of the subsets of the set of variables, - defining marginals for the first cluster alpha, the second cluster beta and the third cluster gamma, respectively; - defining a free energy Fpsi(p) being a functional over the marginals for the first cluster alpha, the second cluster beta and the third cluster gamma, - performing a minimisation of the free energy Fpsi(p) under a normalisation and consistency constraint, wherein performing the minimisation of the free energy comprises performing an inner-loop computation and an outer-loop computation, which computations are consecutively repeated until a first convergence criterion is fulfilled.