In der vorliegenden Arbeit werden verschiedene Reinforcement Learning-Algorithmen und Arten von Classifiern getestet und verglichen. Für den Vergleich werden die ausgewählten Algorithmen mit ihren jeweiligen Classifiern für ein gegebenes Problem einzeln optimiert, trainiert und später im direkten Vergleich gegenübergestellt. Das gegebene Problem ist eine Variante des klassischen Spiels "Tron" . Ein von den Regeln her Einfaches, aber trotzdem bzgl. des Zustandsraumes hochdimensionales Problem und dynamisch in der Entscheidungsfindung. Als Algorithmen werden REINFORCE mit baseline, Q-Learning, DQN und der A3C-Algorithmus ausgewählt. Als Classifier werden lineare Funktionsannäherungen und Convolutional Neural Networks verglichen.
The extraction of individual reference strings from the reference section of scientific publications is an important step in the citation extraction pipeline. Current approaches divide this task into two steps by first detecting the reference section areas and then grouping the text lines in such areas into reference strings. We propose a classification model that considers every line in a publication as a potential part of a reference string. By applying line-based conditional random fields rather than constructing the graphical model based on individual words, dependencies and patterns that are typical in reference sections provide strong features while the overall complexity of the model is reduced.
Terrabytes of geospatial data have been made freely available recently on the Web. For example, data from gazetteers such as Geonames, maps from geospatial search engines like Google Maps and OpenStreetMap, and user-contributed content form social networks such as Foursquare.
Dempster-Shafer is a popular evidence theory that allows for plausible reasoning with belief values, which are associated with a specific item of information. It creates a belief system for reasoning with the uncertain of the information. It may be applied to ontologies, for resolving inconsistencies that were introduced by new added assertions. For instance confidence values that are associated with assertions, may serve as evidences, ie belief values, and the belief system can be applied for resolving such an inconsistency
Every city that incorporates the water-element in its fabric is confronted with the fundamental requirement of developing policies for driving development in the surrounding area, while balancing between economic growth, protection of the environmental, and safeguarding social cohesion.
Pointing is an every-day computer interaction, but a challenge for gaze-controlled input mechanisms. Multiple factors as precision, accuracy and the conflict between the eye used both as sensor and controller limit efficiency and usability. I will present the state-of-the-art; our current approach of continuous zooming, which is enriched by the novel aspects of center offset and deviation; and introduce my idea to detect and fix potential accuracy drawbacks of an eye tracker calibration.
ResearchSpace is an extensible collaborative research environment based on Linked Data and knowledge representation using CIDOC-CRM to provide the context and meaning required for scholarly knowledge building. ResearchSpace is implemented based on the metaphactory, metaphacts' end-to-end platform for creating and utilizing knowledge graphs.
The study of complex networks has received much attention over the past few decades, presenting a simple, yet efficient means of modelling and understanding complex systems. The majority of network science literature focuses on simple one-mode networks. In the real world, however, we often find systems that are best represented by bipartite networks that are commonly analysed by examination of their one-mode projection. One-mode projections, however, are naturally very dense and noisy networks and hence the most relevant information may be hidden. In this talk I present the motivation of my PhD thesis and summarise the research that I conducted during my candidature.
The Linked Data best practices for data publishing encourage the use of RDF to describe URI-identified resources on the Web. As those resources reflect things in the real world, which is without a doubt dynamic, the dynamics of Linked Data should not be neglected. In this talk I report on experimental work on dynamic Linked Data that is based on the Dynamic Linked Data Observatory, a long-term data collection of Linked Data on the Web. Moreover, I cover formal work to capture the dynamics of Linked Data with the aim to specify agents on the Linked Data web using rules. Last, I showcase applications based on the talk topics from the area of cyber-physical systems and the Web of Things.
In the last years, scalable RDF stores in the cloud have been developed, where graph data is distributed over compute and storage nodes for scaling efforts of query processing and memory needs. One main challenge in these RDF stores is the data placement strategy that can be formalized in terms of graph covers. These graph covers determine whether (a) the triples distribution is well-balanced over all storage nodes (storage balance) (b) different query results may be computed on several compute nodes in parallel (vertical parallelization) and (c) individual query results can be produced only from triples assigned to few - ideally one - storage node (horizontal containment).