Adaptive processing of structural data: from sequences to trees and beyond [Elektronische Ressource] / Andreas Küchler, Andreas

Abteilung NeuroinformatikProf. Dr. Gun ther PalmAdaptive Processing of Structural DataFrom Sequences to Trees and BeyondDissertation zur Erlangung des DoktorgradesDr.rer.nat. der Fakult at fur Informatik der Universit at UlmvonAndreas Kuc hleraus Roth1999Amtierender Dekan: Prof. Dr. Uwe Sch oningErster Gutachter: Prof. Dr. Gun ther PalmZweiter Gutachter: Prof. Dr. Uwe Sch oningExterner Gutachter: Prof. Dr. Marco GoriTag der Promotion: 15.02.2000AbstractThere are several challenging real-world problems, for instance in the eldsof medical and technical diagnosis, molecular biology, or document and imageprocessing, where the objects of interest are signi c antly structured, and theircomponent parts have a continuous nature and are subjected to di eren t typesof noise. For a number of practical tasks, the solution can be described byexample data, but there is no or only partial and uncertain prior expert knowl-edge aboutthe relevant structural concepts. Thus, it would beadvantageous tohave methods and tools to (automatically) infer the solutions, i.e. the desiredinput-output mapping, from the given example data.In computer science, structural (for causal, topological, or hierar-chical) relationships between parts of a object are commonly represented bysymbolic formalisms such as graphs, terms or diagrams. Symbolic machinelearning approaches can deal with these representations, but fail if the rangeof the intended mapping is of continuous nature.