PROFILE stands for "Probabilistic First-Order Learning" and is a set of software tools for Statistical Relational Learning and Probabilistic ILP. All software and tools provided here are for academic use only ! Of course, the site and tools are far from being complete. Any helpful comments and informations are welcomed. Please, do not hesitate to contact Kristian Kersting.
Statistical Relational Learning is an emerging field that
| deals with machine learning in relational domains where observations may be missing, partially observed, and/or noisy. |
Traditionally, relational and logical reasoning, probabilistic and statistical reasoning, and machine learning are research fields in their own rights. Nowadays, they are becoming increasingly intertwined. Applications within e.g. bioinformatics, transportation systems, social network analysis, citation analysis, and robotics provide uncertain information about varying numbers of entities and relationships among the entities. Traditional machine learning approaches are able to cope either with uncertainty or with relational representations but typically not with both.
Whereas many of the existing SRL approaches start from a statistical learning perspective and extend probabilistic formalisms with relational aspects, most systems here take a different perspective, in which they start from inductive logic programming (ILP). ILP is the intersection of machine learning and logic programming. It aims at a formal framework as well as practical algorithms for inductively learning relational descriptions (in the form of logic programs) from examples and background knowledge. It does not, however, explicitly deal with uncertainty. Most tools here extend ILP formalisms, settings and techniques to deal with probabilities.