Search improvements in multirelational learning [Elektronische Ressource] / von Maria de Lourdes Peña Castillo

Search Improvements inMultirelational LearningDissertationzur Erlangung des akademischen GradesDoktoringenieurin(Dr.-Ing.)angenommen durch der Fakultat fur Informatik der Otto-von-Guericke-Universitat Magdeburgvon M. Sc. Mara de Lourdes Pen˜a Castillogeboren am 02.Mai1974 in Mexico, D.F.Gutachter: Prof.Dr. Stefan WrobelProf.Dr. Stefan KramerProf.Dr. Andreas NurnbergerMagdeburg, den 7. June 2004AbstractIn this thesis we lay the foundations to develop multirelational learning systemswhich can cope better with the challenges posed by structural and topologicaldomains. Even though many interesting application domains contain structuralor topological data, current multirelational systems have diculties dealing withthe complexity of the search space of theses domains, their indeterminacy, andthe presence of non-discriminating relations. In this work, we describe macro-operators which are a formal method to reduce the search space explored instructural or topological domains and can also be used to alleviate the myopiaof greedy systems. We also explore parallel search based on stochastically se-lected examples to reduce the instability of example-driven learning. As a thirdcontribution, we present active inductive learning as an approach to improve thee ciency of the instance space exploration.