@MISC{Cussens_ongenerative, author = {James Cussens and P (x Z}, title = {On Generative Parameterisations of Markov logic networks}, year = {} }

Share

OpenURL

Abstract

Given some fixed first-order language, a Markov logic network (MLN) (Domingos et al., 2008) uses weighted formulae to define a probability distribution over Herbrand interpretations of the language. (To save space a Herbrand interpretation will be called a ‘world’ from now on. In this paper all such formulae will be clauses.) “An MLN can be viewed as a template for constructing Markov networks. ” (Domingos et al., 2008) and many analyses of MLNs has taken Markov networks as their point of departure. Here MLNs are analysed from the more general perspective of exponential-family distributions. An MLN is a set L = {(Fi, wi)}i=1,...k where each Fi is a first-order clause and wi is the weight associated with that clause. For any world x, the MLN assigns it the following probability: