Closed-Form Solutions in Learning Probabilistic Logic Programs by Exact Score Maximization

Francisco Henrique Otte Vieira de Faria, Fábio Gagliardi Cozman, Denis Deratani Mauá

Abstract: We present an algorithm that learns acyclic propositional probabilistic logic programs from complete data, by adapting techniques from Bayesian network learning. Specifically, we focus on score-based learning and on exact maximum likelihood computations. Our main contribution is to show that by restricting any rule body to contain at most two literals, most needed optimization steps can be solved exactly. We describe experiments indicating that our techniques do produce accurate models from data with reduced numbers of parameters.