Taming the Complexity of Inductive Logic Programming

Filip Zelezný, Ondrej Kuzelka

Abstract: Inductive logic programming (ILP) [12] is concerned with the induction of theories from specific examples and background knowledge, using first-order logic representations for all the three ingredients. In its early days some twenty years ago, ILP was perceived as a means for automatic synthesis of logic programs, i.e. Horn clausal theories. Current research views ILP algorithms mainly in the context of machine learning [14] and data mining [1]. ILP has enriched both of the two fieds significantly by providing them with formalisms and algorithms for learning (or ‘mining’) complex pieces of knowledge from non-trivially structured data such as relational databases.