Version vom 7. Juli 2009, 10:29 Uhr

Events

Document Image Dewarping and Illumination Correction using Reference Templates2024-07-03T15:45:000Graduiertenkolloquium
Termin: 3 Juli 2024 15:45:00
Beschreibung DE: In our partially digitalized world, printed documents are ubiquitous, despite the ongoing efforts to digitalize them.

Printed forms play essential roles in various business workflows, such as tracking orders and goods in commercial shipments. To process the information on these forms, data is usually inputted into a computer system through scanning or manual entry. However, both of these methods are time-consuming and inflexible, as there is not always a scanner available nearby. To mitigate this problem, recent research has focused on the development of data extraction systems that can automatically extract data from images of printed forms captured by smartphones. Due to unrestricted environmental conditions and document deformations during the capturing process, the images are often of poor quality, which makes the extraction process difficult. More specifically, the environmental factors include ambient lighting and shadows, whereas document deformations result from the capturing angle and the document's physical condition, such as bends, crumples, folds, and similar factors. In order to improve the extraction process, the images are often enhanced by geometric dewarping and illumination correction. The former aims to remove document deformations, while the latter aims to remove the effects of uneven lighting.

In contrast to prior work, we integrate reference template images in the document enhancement process. These reference templates are RGB images of the document in its digital version but without any information on the document's content, thus providing information about the expected layout. The layout information includes the document's structure, such as the position of the texts, structural lines and, logos, as well as the visual appearance of the aforementioned elements. By leveraging the prior knowledge of the document's structure and visual appearance, we improve the geometric dewarping and illumination correction processes. In particular, we propose two approaches for geometric dewarping and one approach for illumination correction, all of which integrate additional reference template information. Thereby, we bring the document images closer to their digital counterparts, which in turn improves the performance of the subsequent data extraction process.

Im Rahmen der Arbeit wurde anhand theoriebasierter und empirischer Ansätze ein umfassendes Kompetenzmodell entwickelt, das Lernziele für zentrale Handlungsbereiche der grafischen Modellierung und überfachliche Kompetenzen beschreibt. Es wurde ein Aufgabenkatalog erstellt, der Aufgabentypen mit den im Kompetenzmodell definierten Lernzielen verknüpft. Ergänzend wurden exemplarische Bewertungsschemata und Empfehlungen für die Gestaltung lernförderlicher Feedback-Nachrichten auf Basis des Kompetenzmodells abgeleitet. Die Ergebnisse unterstützen Lehrende bei der Auswahl von Lernzielen und der Gestaltung kompetenzorientierter Prüfungen anhand passender Modellierungsaufgaben.

Zur Umsetzung kompetenzorientierter Prüfungen auf dem Gebiet der grafischen Modellierung wurde eine E-Assessment-Plattform entwickelt. Diese berücksichtigt verschiedene grafische Modellierungssprachen, individuelle Bewertungsschemata und Feedback-Empfehlungen. Zusätzlich wurden Dienste zur automatisierten Bewertung von Petri-Netzen erstellt, die Lernziele zu syntaktischen, semantischen und pragmatischen Qualitätsaspekten adressieren. Die Einsatzfähigkeit der Plattform wurde im praktischen Einsatz in Lehrveranstaltungen und Prüfungen demonstriert. Ergänzend wurden Befragungen zur Benutzungsfreundlichkeit und weiteren Aspekten durchgeführt und die Ergebnisse der Anwendung der Bewertungsdienste auf einer umfangreichen Datenbasis studentischer Petri-Netze evaluiert.

To address these issues, we present a set of data representation and information extraction approaches that enable the creation of machine-readable publication corpora more extensive, more complete, less noisy, and of finer granularity than previously available. In particular, we present the following. As the foundation of our research, we introduce unarXive, a large-scale data set comprising the full-text and citation network of over 1 million publications. Utilizing unarXive, we further present approaches yielding advances in three areas. First, we demonstrate improvements of the completeness of citation networks though the use of blocking, as well as improvements to the granularity of document representations. Second, we show advances in the language coverage of document interconnections through a large-scale analysis of cross-lingual citations. Third, we present information extraction approaches for fine-granular representations of research artifact citations. Overall, our approaches address key shortcom-ings of currently used machine-readable representations of scientific publication corpora. With our approaches we demonstrate their benefits through evaluations and practical large-scale applications. Sig-nificant parts of our work have already seen adoption within the research community, which confirms their utility.

This thesis focuses on addressing these two challenges in Knowledge Graph Completion: • Predicting the missing links between entities in KG via head, and tail prediction as well as triple classification. • Predicting the semantic type of an entity in a KG.

To date, many algorithms are proposed to learn the embeddings of the entities and relations into the same vector space. However, none of the baseline models considers the contextual information of the KGs along with the textual entity descriptions to learn the latent representation of the entities and relations within the KG. This thesis proposes a multi-hop attentive KG embedding model MADLINK for predicting the missing links between the entities.

On the other hand, recent research on entity typing focuses on exploiting the structural information of the KGs, and keywords from the entity description for the prediction task. A novel entity typing model CAT2Type, to predict the missing types of the entities in a KG leveraging the Wikipedia Category information, is proposed in this thesis. Furthermore, to predict the multi-label, multi-class, and hierarchical entity types in KG, a GRAND framework is proposed that exploits graph walks and entity descriptions. The results show that the proposed methods, MADLINK, CAT2Type and GRAND outperform the current state-of-the-art models for the benchmark datasets in the corresponding tasks.

In this work, we address those explanation goals with four conceptual approaches. First, we present FAIR-nets, which aims to process metadata such as intended use and architecture information in order to make neural networks more transparent and efficient. Second, to open the neural network box, we define a new explanation quality metric ObAlEx for image classification. Using object detection approaches, explanation approaches, and ObAlEx, we quantify the focus of CNNs on the actual evidence which covers scrutability, trust, and effectiveness. Third, we propose FilTag, an approach to explain Convolutional Neural Networks even to non-experts by tagging the filters with keywords. These tags provide an explanation of what the filter does. Individual image classifications can then be intuitively explained in terms of the tags of the filters that the input image activates. These explanations enhance scrutability and trust. Last but not least, we present TransPer, an explanation framework for recommender systems based on neural networks. We defi-ne explanation measures based on Layer-Wise Relevance Propagation to understand the recommendation quality and find new ideas on how to improve the recommender system. This captures transparency, trust, persuasiveness, and satisfaction.

The given talk will focus on approaches FAIRnets, ObAlEx, and TransPer to showcase our framework and give insights into developed explanation techniques for neural networks.

Still, today’s stream processing pipelines are focused on static deployments in centralized cloud environments that come with inevitable downsides, particularly in the context of IoT scenarios that require fast results, are limited by the available bandwidth, or bound to regulations regarding data sovereignty and ownership. In recent years, a new decentralized computing paradigm referred to as Fog Computing allows to overcome the shortcomings of previous cloud-only approaches by offloading certain processing in closer proximity to the data source. This offers possibilities for a next generation of stream processing applications that are centrally modeled, and deployed over a pool of geographically distributed compute resources. However, unlike the cloud, fog and edge nodes are highly heterogeneous in terms of hardware specifications and are potentially exposed to the physical world. Deployment-wise, current systems still require deep technical knowledge in infrastructure management and lack an approach for explicitly exploiting heterogeneous computational resources available in the underlying fog infrastructure allowing to autonomously select suitable deployment configurations in order to better assist domain experts along the deployment process.

In this presentation, we show how to abstract application specialists from the technical complexities of deploying, executing and adapting stream processing pipelines in heterogeneous fog infrastructures. Our main contributions are (i) a generic node model describing relevant resource characteristics and additional metadata, (ii) a node controller incorporating the node model for pipeline element life cycle management, as well as (iii) a fog cluster management framework to flexibly deploy, refine and relocate individual pipeline elements to registered nodes along the cloud-edge continuum while handling messaging between adjacent elements. We have evaluated resource consumption and deployment metrics of our work in two application domains, manufacturing and urban logistics.

A major challenge that hinders the full adoption of industrial event stream analytics is the distribution of required technical and domain knowledge among several people. This makes the realization of analyt-ics projects time-consuming and error-prone. For instance, accessing industrial data sources requires a high level of technical skills due to a large heterogeneity of protocols and formats and long-lasting operating times of industrial assets. In addition, there is a gap between domain knowledge required to understand complex production processes and data science/operations knowledge to implement ana-lytics applications based on high-frequency industrial event streams. Altogether, there is a lack of a holistic approach that focuses on domain experts.

To solve these challenges, a new set of methods and tools is required that covers the entire data ana-lytics life cycle for industrial event streams. Data must be connected and analyzed (e.g. train machine learning models). These models must then be deployed in a geographically distributed architecture. There exist specialized solutions for the individual tasks and the combination of all these tasks can become rather complex.

In this work, we enable domain experts to exploit the full potential of data-driven decision making in industry scenarios. Therefore, we present an end-to-end framework to connect, transform and analyze industrial event streams without deep technical expertise. We leverage semantic models that are used to automatically instantiate adapters at the edge and extract preprocessing rules. Subsequently, these models are used i) to automatically adapt event streams (e.g. frequency, quality) according to pro-cessing requirements at the edge and ii) to better exploit event streams by enabling domain experts to label data in an intuitive way and train new machine learning models automatically. We provide an edge-based semantic adapter library to accelerate data connection. Connected data streams are dynamically adapted according to the requirements of the algorithms processing the events. Domain experts can use a labeling editor to label data and train machine learning models for time-series data using an automated machine learning service. We have evaluated both the performance of our solutions and their applicability and usability in a conducted user study. The proposed methods are accompanied by extensive tool support, which is fully available as open source.

In this presentation, we show how our work enables domain experts to leverage machine data and aim to reduce the effort of data analytics projects on industrial event streams.

Graph structured data are ubiquitous data structures, used to model relationships between entities. Graphs have become an important foundation to represent interactions between users in social networks, items in recommender systems, and interactions between drugs in bioinformatics. The main research problems in these areas include node clustering, node classification and link prediction. Especially the link prediction task is in bioinformatics of special interest toward the identification and development of new uses of existing or abandoned drugs since drug development is currently very time consuming and expensive.

In recent years, network embeddings are often trained for encoding the nodes of a network into a low-dimensional space whilst preserving the graph structure. Based on the trained embeddings, machine learning techniques are applied to address the main machine learning tasks, such as link prediction. In most of the published methods, like e.g. DeepWalk, node2vec and LINE, random walks procedures are used to efficiently explore diverse neighbourhoods and compute embeddings based on them. However, these methods assume a connected graph, otherwise the random walks are not sufficient to gather enough information of nodes in the neighbourhoods.

In this work we address three types of problems: Link prediction on bipartite networks, link prediction on knowledge graphs and a semantic grouping of nodes and links in graphs. We use a stochastic factorization model to learn a target distribution over the graph structured data, allowing to predict unknown links and embed the nodes into a low-dimensional space whilst preserving the distribution of interactions within the graph. Compared to the existing methods that use random walks, our approach is much more robust in terms of the connectivity of the networks. Results show that the proposed method outperforms current state-of-the-art models in several studied complex networks and sets a new baseline in link prediction on disconnected graph structured data.

For this presentation we focus on the first two research problems, namely link prediction on bipartite networks and link prediction on knowledge graphs. We show how a stochastic factorization model is used to learn a target distribution on the graph structured data, based on the learned target distribution predict missing links and the benefit of our method in terms of the connectivity of the graph structured data.

Thus, is there a way to develop better and more resilient models and simulations that will update themselves along with the real systems? Even more, can we have models being directly extracted from real systems? How can we minimize the dependence on expert knowledge and is that even possible? To what extent?

The talk will address the use of M&S for analysis of Cyber-Physical Systems, with a special focus on Reliability Modeling and the use of data to enhance M&S processes.

Sanja Lazarova-Molnar is an Associate Professor with the Faculty of Engineering at the University of Southern Denmark, researching in the areas of Modeling and Simulation, and Data Analytics, in various decision support contexts. Sanja is a Senior Member of IEEE, and currently serving as Director-at-Large on the Board of Directors of The Modeling & Simulation International (SCS). Furthermore, she is Chair of IEEE Denmark Women in Engineering Affinity Group. She is also one of the Proceedings Editors for the Winter Simulation Conference 2019. As of latest, her focus is on the data-based reliability modeling and analysis of cyber-physical systems, such as manufacturing systems and smart buildings. Sanja completed her PhD in 2005 at the University in Magdeburg, specializing in the area of Modeling and Simulation. During her PhD, she developed and formalized new simulation method, i.e. the proxel-based method. The proxel-based method was implemented in the DaimlerChrysler simulation tool and it was superior to discrete-event simulation for certain classes of problems, such as warranty analysis. Sanja is author of more than 90 papers published in international conferences, books and journals

Im Rahmen des Vortrags wird eine Methodik vorgestellt, die ein heterogenes Energiesystem eines Automobilwerkes bestehend aus einer Vielzahl an steuerbaren und nicht-steuerbaren Erzeugern, -verbrauchern und -speichern untersucht. Es wird ein Optimierungsproblem entwickelt, das einen Einsatzplan der steuerbaren Energieanlagen als Ergebnis hat, mit dem die notwendigen Energie-bedarfe kostenminimal gedeckt werden und zusätzlich Erlöse durch das externe Angebot von Flexibilität erzielt werden können. Das Optimierungsproblem wird als gemischt-ganzzahliges lineares Programm modelliert und exakt gelöst.

Bio: Prof. Jens Grossklags, Ph.D., holds the Associate Professorship for Cyber Trust at the Department of Informatics at the Technical University of Munich. He studies security and privacy challenges from the economic and behavioral perspectives with a variety of methodologies. His published research has appeared in high quality publications at technical and social science venues, and has been honored with three best paper awards. Prof. Grossklags received his Ph.D. from the University of California, Berkeley and was a Postdoctoral Research Associate at the Center for Information Technology Policy at Princeton University. He then directed the Security, Privacy and Information Economics Lab, and served as the Haile Family Early Career Professor at the Pennsylvania State University. In addition, Prof. Grossklags has been an invited visiting professor at École Polytechnique Fédérale de Lausanne (EPFL), Switzerland, during the summers of 2011 and 2012, an invited visiting researcher at Copenhagen Business School (CBS) during the summer of 2013, an invited visiting scientist at EURECOM during the summer of 2014, and an invited professor at the IMDEA Software Institute during the summer of 2016.

In diesem Vortrag wird eine Methode vorgestellt, die bei der Auswahl geeigneter Maßnahmen zum Schutz der Bevölkerung hilft. Die Arbeit ist Teil der Vorsorgeforschung mit dem Fokus auf kerntechnische Unfälle. Die entwickelte Methode berücksichtigt (i) das Problem der Unsicherheit in der Entscheidungsfindung, (ii) die Einzigartigkeit der betrachteten Ereignisse, (iii) die strukturierte Integration von Erfahrung und Expertenwissen, (iv) die Durchführungsreihenfolge der Maßnahmen, (v) die Integration von Akteuren mit unterschiedlichen Präferenzen und (vi) die Möglichkeiten der Anwendung mit Hilfe eines computergestützten Werkzeugs. Die Methode erweitert insbesondere das Fallbasierte Schließen um die Entwicklung geeigneter Szenarien, ein auf Petri-Netzen basierendes Modell und eine multikriterielle Bewertung mehrerer Handlungsalternativen. Eine prototypische Implementierung zeigt die Anwendbarkeit der entwickelten Methode auf.

The presentations will take place on Tuesday, April 16, 2019:

title: Design and Prototypical Evaluation of a Blockchain-Based System for the Storage of Electronic Medical Records

presenter: Patrick Frey

time: 15:00 Uhr

place: room 1A-11

abstract: The demographic change poses significant challenges to the healthcare ecosystem. One challenge is the rising average age of the world's population. This results in a larger number of patients. To counteract this trend, it is important to maintain or even improve the quality of individual care while handling a higher number of patients. This improvement can be supported by providing access to the patients' Electronic Medical Records (EMR). However, these records are only available in fragmented form on data silos of healthcare providers. The objective of this study is to build a system design which creates a common data basis and thus dissolves the existing data silos. With the use of Blockchain Technology (BCT) and a distributed file system, a unique possibility is offered to support a trustful interaction in an ecosystem by providing characteristics like decentralization, transparency, immutability, and pseudonymity. A decentralized application controls the interaction of the components and acts as an interface for the user. Through the user-friendly interface, EMR can be stored as well as accessed by stakeholders in the healthcare ecosystem. For the creation of the system design, requirements are recorded and a suitable BCT protocol is selected. The proposed system design provides a common data basis that ensures privacy, security, reliability, and accountability. It enables a faster processing time for EMR sharing and improves decision making for stakeholders. A prototype is implemented demonstrating the technical feasibility of the system design.

title: A Qualitative Analysis of GDPR-related Issues and Compliance Practices in German Public Offices

presenter: Stefan Mill

time: 15:30 Uhr

place: room 1A-11

abstract: Data protection is one of the central topics of our society as the digitalization of our daily lives proceeds. However, discussions about data protection in theory and practice often only refer to the technological giants and social networks of our time. The public sector, its issues and practices with regard to data protection is widely ignored. Even the research community is rather restrained with respect to this specific field. Even with the considerable peak of interest in May 2018, caused by the GDPR, the data protection within the public sector, again, did not benefit.

This master thesis focuses on this lack in research. Using an exploratory approach, interviews with data protection officers of municipal public offices has been conducted in the months following the introduction of the GDPR. The sample comprises a total of 18 data sets, collected from cities from of all sizes all states of Germany. Based on this study, the thesis summarizes the main statements, issues and solution approaches with regard to general data protection and the implementation of the GDPR in four dimensions: organization, culture, context, and legislation. Using the central insights of every dimension, it discusses the main relations of themes as well as the relevance of personnel resources and the data protection practices in the past. The thesis provides suggestions for practitioners and works out potential areas for further research for fellow scientists.

Bio

Michael Cochez, is a postdoctoral researcher at the Fraunhofer Institute for Applied Information Technology FIT in Germany. In this position Michael is working on transferring research results from the academic world to the industry. Besides the industry exposure, he conducts research in areas related to data analysis and knowledge representation, like knowledge graph embedding, scalable clustering, frequent itemset mining, stream sampling, prototype-based ontologies, ontology matching, and knowledge evolution. This research is currently mainly conducted at the RWTH Aachen university, Germany. Before joining Fraunhofer, he obtained his Ph.D. degree from the University of Jyväskyä, Finland under the supervision of Vagan Terziyan and Ferrante Neri (De Montfort University - Leicester). He obtained his master degree from the same university and his bachelor degree from the University of Antwerp, Belgium. Michael Cochez is currently on a partial leave from a postdoc at the University of Jyväskylä and is also a scientific advisor for WE-OPT-IT Oy (former MyOpt Oy) in Finland.

For this reason, this work aims for providing a framework that allows the formalisation of use case specific tasks and to simulate/imitate them in a general way so that agents are enabled to train their inherent characteristics and apply the resulting ML models for solving tasks adequately and domain independent.

In this talk, a machine learning algorithm is presented that was originally designed to simulate a part of human thinking itself, namely associating: the process of navigating from one thought to the next. The presented approach is an evolutionary algorithm which can learn graph patterns (SPARQL queries) from a given SPARQL endpoint based on a given list of training source-target entity pairs. After training, given a new source node, the algorithm can predict target nodes analo-gously to the training list.

It will be presented, how the Graph Pattern Learner operates in an end-to-end learning fashion and extracts features in form of graph patterns without the need for human intervention from huge, noisy, real world knowledge graphs. Further, it will be shown, how the learned graph patterns form a feature space (embedding) adapted to the given list of examples, and being SPARQL queries, can be explained and executed to generate target candidates. The talk will conclude with current research areas arising from these properties.

The second part of this talk dives into a combination of Knowledge Graphs and Deep Learning methods for NLP tasks, i.e., it discusses the details of Framester, a linguistic linked data hub with a recent addition of MetaNet (a resource for Metaphors). This resource is further used for generating Frame/Role Embeddings for knowledge reconciliation over the knowledge graphs generated from text.

The third part of this talk gives a vision on other Natural Language Processing tasks that can be performed using the previously defined resources with Deep Learning Techniques such as semantic textual similarity, refining MetaNet metaphors, metaphor generation, metaphor detection/interpretation etc.

Our hypothesis is that by fusing the representations, we attain a more holistic representation for identifying similarities as the modalities cover different aspects of an entity, e.g. visual attributes of entities cover shape and color information that is not easily covered in other modalities. How-ever, this fusion is limited to concepts with cross-modal alignments in the training data which are only available for a few concepts. Since alignments over different modalities are rare and expensive to create, we finally investigate an extrapolation approach to translate entity representations outside of the training corpus to the shared representation space.

However, searching for additional learning material on the Web might lead to distraction from the learning task and to abandonment of the assessment eventually.

Embedding appropriate Web resources on a technical and semantic level might provide a solution to this problem. In order to do so, we need to provide metadata about the content as well as of educational characteristics. Such data is essential to enable integration and interoperability of rele-vant material in the right context. Furthermore, we have to deliver that material in a format that makes Web-based integration also technically feasible.

In this talk we propose an approach that transforms general Web resources into so-called Linked Learning Items (LLI). These Linked Learning Items can be integrated into learning management systems (LMS) and serve as additional learning resources available to learners for studying and assisting with assignments.

Das Institut AIFB und der Verein AIK e.V. laden Sie nun sehr herzlich zu folgendem Vortrag ein:

Prof. Dr. Ali Sunyaev (Institut AIFB, KIT):
„Sorgenfrei in die Zukunft dank Trusted Blockchain –
Was ist DLT? Was ist Blockchain? Warum betrifft das mein Unternehmen?“

Im ersten Teil des Vortrags werden zunächst Ansätze vorgestellt, die eine Aufteilung des Gesamtenergieverbrauchs in den Verbrauch einzelner Haushaltsgeräte (Non-Intrusive Load Monitoring Verfahren, NILM) mit Hilfe von tiefen neuronalen Netzen (Deep Learning) realisieren. Dabei wird ein neuer Ansatz vorgestellt, generative neuronale Modelle einzusetzen, um die Produktion von Gerätelastprofilen zu unterstützen. Erste Ergebnisse mit Generative Adversarial Networks (GAN) als generatives Modell werden präsentiert. Der zweite Teil des Vortrags analysiert, in wie fern NILM als Datenanalyse durch eine Reduktion der zeitlichen Auflösung behindert werden kann. Dies liefert erste Indizien, ob ein Energieanbieter mit der aktuellen Gesetzeslage (insbesondere mit den aktuellen technischen Richtlinien des BSI zum Smart Meter Gateway) in der Lage wäre, Einblick in die Aktivitäten im Haushalt zu erlangen.

Schließlich wird im dritten Teil des Vortrags mit der Anwendung von Differential Privacy auf Smart-Metering-Daten, die zum Zweck einer Energieverbrauchsvorhersage beim Energieversorger erhoben worden sind, ein Privatsphärenkonzept vorgestellt. Dabei werden die Faktoren Nützlichkeit für den Datennutzer und Privatsphärenverlust des Betroffenen evaluiert. Der betroffene Energiekunde erhält damit eine Garantie für die obere Grenze des Risikos eines Privatsphärenbruchs durch den Energieanbieter oder Dritte.

The user interface of a building operating system in a smart home plays an essential role in realiz-ing demand response, since it helps to increase the residents' energy-awareness and facilitates load shifting by allowing residents to configure degrees of freedom for their appliances. This talk will present a generic user interface for building operating systems from aspects of design, imple-mentation and evaluation.

To ensure the user interface can be flexibly adapted to various types of buildings, a series of ge-neric data models is designed which are independent of any building operating system. Besides, we introduce three roles with different permissions and several functional components of the user interface. Based on this design, a prototype of such a generic user interface named Building Op-erating System User Interface (BOS UI) has been implemented to operate the Energy Smart Home Lab (ESHL) at the Karlsruhe Institute of Technology (KIT). Finally, we evaluate the design, functionality, and usability of the BOS UI. The detailed functions that have been implemented in the BOS UI as well as the evaluation results about the user interface will be presented in this talk.

Der Vortrag stellt ein System zur Online-Optimierung der Homogenisierungsanlagen vor. Die Optimierung wird basierend auf den konkurrierenden Kriterien Homogenisierungsgrad, Effizienz und Robustheit durchgeführt. Der Entscheidungsträger kann mit den aktuellen Analysen und Vorhersagen des Systems eine fundierte Entscheidung zum optimalen Betrieb treffen.

Durch Online-Überwachung der Material-Parameter und die Nutzung der bereits in den Werken verfügbaren Informationen zum Material-Zeitplan ist die kostengünstige Nachrüstung des Optimierungssystems in bestehende Anlagen ohne aufwändigen Umbau möglich. Dadurch können existierende Anlagen dynamisch gesteuert, überwacht und effizient betrieben werden. Im Analysebetrieb bietet das System Möglichkeiten zur frühzeitigen Erkennung zukünftiger Probleme und ermöglicht so eine rechtzeitige Reaktion.

Im Rahmen der Forschung wurden problemspezifische sowie generische Modelle und Algorithmen entwickelt. Darauf basierend wurde Simulations- und Optimierungs-Software implementiert. Diese werden zusammen mit den erzielten Ergebnissen und der Evaluation der entwickelten Komponenten in diesem Vortrag vorgestellt.

Eine Ausprägung hiervon sind intelligente Gebäude, die mittels eines automatisierten Gebäudeenergiemanagementsystems die Energieflüsse in Haushalten durch Ausnutzung von Speichern und von Lastverschiebungspotentialen weiterer lokaler Komponenten gemäß nutzergegebener Präferenzen optimieren. Die Optimierungsziele sind zumeist ökonomischer Natur und vom Netz-betrieb entkoppelt. Daher können die Ziele der Nutzer des Gebäudes mit denen des Betreibers des versorgenden Niederspannungsnetzes in Konflikt stehen, da der Netzbetreiber einen zuverlässigen Netzbetrieb sicherzustellen und die dafür nötigen Handlungen zu ergreifen hat. Diese Eingriffe können auch in die Sphäre von intelligenten Gebäuden reichen und deren Handlungsspielraum einschränken.

Der Vortrag analysiert die Auswirkungen intelligenter Gebäude auf Niederspannungsnetze und diskutiert die Anforderungen an ein regionales Energiemanagementsystem zur netzdienlichen Koordination von intelligenten Gebäuden. Ein entsprechendes System wird vorgestellt, in dem eine Netzampel zur Auflösung des skizzierten Interessenkonflikts zwischen Netzbetrieb und Netznutzung ausgestaltet ist. Weiterhin wird für die Technologieevaluation von regionalem Energiemanagement und den Netzrückwirkungen intelligenter Gebäude eine Power-Hardware-in-the-Loop Umgebung aus einem realen intelligenten Gebäude und einer Microgrid-Co-Simulation auf Basis definierter Referenzszenarien vorgestellt.

This talk will address the involvement of data consumers in the process of data cleans-ing and issues of handling dynamic Linked Data.

Linked Data publishers can apply various data quality evaluations prior to publication of their datasets. But nevertheless, most inconsistencies only become apparent when the data is processed in applications and presented to the end users. Therefore, it is not only the responsibility of the original data publishers to keep their data tidy, but progresses to become a mission for all distributors and consumers of Linked Data. Hence, data publi-cations ideally get continuously updated in order to match the quality needs of the actual data consumers. My main research topic is the inspection of feedback mechanisms for Linked Data cleansing in open knowledge bases. This work includes processing change requests created by data consumers, versioning Linked Data resources, and methods for individual user notification about data updates.

Im Rahmen dieses Vortrages präsentieren wir eine konzeptionelle und technische Erweiterung für Complex Event Processing Systeme zur transparenten und nachvollziehbaren Verarbeitung von Datenströmen. Hierzu wird in einem ersten Schritt der implizite Lebenszyklus für Situationen erweitert und explizit eingeführt. Grundlage der Erweiterungen bildet das psychologische Modell des Situationsbewusstseins, das beschreibt wie Menschen im Kontext von komplexen Systemen Informationen verarbeiten. Technisch basieren die notwendigen Erweiterungen der Architektur und Arbeitsweise von Complex Event Processing auf Basis der teilweisen Musterausführung (Partial Pattern Fulfillment). Dem Anwender wird somit die Möglichkeit gegeben, die Entwicklung einer Situation zur Laufzeit nachzuverfolgen, relevante Statistiken einzusehen und im Idealfall die Situation sogar vorherzusagen. Diese neugewonnene Transparenz kann somit genutzt werden, um die Weiterentwicklung von Complex Event Processing von einer rein reaktiven Sicht hin zu einer proaktiven Sicht voranzutreiben.

Im Vortrag wird ein Ansatz zur Minimierung der Betriebskosten von Energiesystemen in Gebäuden unter Unsicherheit vorgestellt. Basierend auf der Optimierung eines rollierenden Planungshorizontes wird der Einsatz von elektrischen und thermischen Energiespeichern, eines Blockheizkraftwerkes sowie verschiebbarer Lasten geplant. Hierbei werden die Unsicherheit der prognostizierten Erzeugung der Photovoltaikanlage in der Optimierung durch stochastische Methoden, die Problematik unterschiedlicher Zeitskalen und die Wahl des Planungshorizontes berücksichtigt. Der Vortrag präsentiert das Modell des Energiesystems, die Optimierungsmethodik und Vergleiche mit deterministischen Ansätzen.

In this work, we develop such forecaster investigating nonparametric regression methods such as Kernel and K-Nearest Neighbors regression. We demonstrate its accuracy on a large set of low voltage consumers of different types and aggregation size, comparing them to load profiling, similar day, as well as other approaches discussed in the literature.

Ein weiteres Problem besteht in der meist nachrangigen Betrachtung von Qualitätsanforderungen im Entwicklungsprozess. Obwohl gerade die Erfüllung der Qualitätsanforderungen über einen erfolgreichen Einsatz des Informationssystems entscheidet, werden Qualitätsanforderungen häufig nicht in die Entwicklungsmodelle integriert und somit auch nicht in auf diesen Modellen basierende Entwicklungsentscheidungen einbezogen. Notwendig ist folglich, dass Qualitätsanforderungen in gleicher Weise wie funktionale Anforderungen mit den Unternehmenszielen abgestimmt sein müssen und systematisch daraus entwickelt werden sollten. Insbesondere bei der Erstellung von Geschäftsprozessmodellen ausgehend von Unternehmenszielen ist es nötig, Qualitätsanforderungen in die Modelle zu integrieren, da die Geschäftsprozessmodelle nicht nur zur Dokumentation dienen, sondern ebenfalls Grundlage für die Ausführung mit Workflowmanagementsystemen sein können. Im Modell integrierte Qualitätsanforderungen können in diesem Fall direkt verstanden und umgesetzt werden.

Im Vortrag wird ein Ansatz beschrieben, der ein Modell der Ebene Unternehmensziele systematisch in ein Geschäftsprozessmodell transformiert, wobei Qualitätsanforderungen in die Transformation einbezogen und in das Geschäftsprozessmodell integriert werden. Eine Anwendung dieses Ansatzes unterstützt ein systematisches und qualitätsorientiertes Vorgehen bei der Entwicklung betrieblicher Informa-tionssysteme.

All Events as Feed:RSS

Colloquium news

You can subscribe to our colloquium mailing list. .
Each announcement will be sent to you by e-mail.
Subscription: Kolloquiumsverteiler

@@ Zeile 13: / Zeile 13: @@
 | link=?Name EN
 }}
+All Events as Feed:{{#ask:[[Kategorie:Veranstaltung]][[Start::+]]
+| ?Start
+| ?Veranstaltungsart=
+| format=RSS
+| searchlabel= RSS
+|rsstitle=Events by AIFB
+|
+}}<br>
 '''Colloquium news'''

Veranstaltungen/en: Unterschied zwischen den Versionen

Version vom 7. Juli 2009, 10:29 Uhr

Events