Algorithmic learning for knowledge-based systems. Pt. 2 Final report

The investigations within the joint research project GOSLER have been focussed on theoretical problems of learning from usually incomplete information (so-called inductive inference). This research has been motivated by the underlying assumption that knowledge-based systems' applications in complex domains are normally faced to the problem of vagueness and incompleteness of information both about the system or installation considered and the dynamic processes maintained. Among the main inference mechanisms of knowledge-based systems, inductive inference will usually play some key role. The joined project yields some considerable contribution to the theory of inductive inference. The research focussed mainly on theoretical results and prototypical implementations. The main emphasis of the research efforts was put on special effects in inductive inference (like inconsistent hypothesis formation, for instance) which are not yet taken into consideration and implemented in most of the recent knowledge-based systems in use. The theoretical results have been validated and applied by building prototypical systems like T_LP"S and GOSLERP. T_LP"S is an interactive system for design and testing of rewrite systems. GOSLERP is based on the logic programming paradigm with the emphasis on certain metaprogramming features (backtracking of assert and retract). GOSLERP is particularly taylored towards the creation of learning modules for knowledge-based systems. (orig.)Available from TIB Hannover: F95B1493+a / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekSIGLEBundesministerium fuer Forschung und Technologie (BMFT), Bonn (Germany)DEGerman