Research Colloquium Cognitive Systems

General Information

The research colloquium of the CogSys group is a forum for scientific discussion between lecturers and doctorate, diploma and master students. Talks cover concepts and results of theses as well as current research in the field cognitive systems. Furthermore, current subjects of more general interest can be presented and discussed. Sometimes external researchers are invited.

If you want to be informed about current talks, please sign up for the course in the virtual campus.

In some semesters, we participate in the Monday Afternon Club (MAC) which is a joint colloquium of the groups of theory (Mendler), programming languages (Lüttgen) and cognitive systems.

Talks WS 12/13

Kolloquium Mittwoch, 27.03.2013, 10:00 Uhr c.t., WE5/5.005

Johannes Wicht (MA AI Präsentation): Entwicklung eines Systems zur Erkennung von Eskalationen in der Assist-Datenbank (in Kooperation mit Siemens)

Ziel der Arbeit war es, ein Programm zu entwickeln, das erkennt, bei welchen Fällen im Customer Support für komplexe technische Anlagen Eskalationspotential vorhanden ist. Dazu werden Methoden aus dem Information Retrieval, wie Data Mining und Text Mining und aus dem Bereich des maschinellen Lernens verwendet. Zunächst wird auf die Abteilung I CS LS SSC der Firma Siemens, für die diese Arbeit erstellt wurde, und die Arbeitsweisen von Hotline-Systemen eingegangen. Anschließend werden das erarbeitete Konzept und die Umsetzung in ein Programm vorgestellt. Schließlich erfolgt eine ausführliche Testung und Bewertung bezogen auf reale Daten aus dem Support.

Kolloquium Montag, 26.11.2012, 16:00 Uhr c.t., WE5/5.005

Matthias Linhardt (MA CiTH Präsentation): Konzeption und exemplarische Ausführung eines
Assistenzsystems für ein Content-Management-System (in Kooperation mit docufy)

Ziel der Arbeit war es, in einem konkreten Anwendungsfall den Nutzer eines bestehenden Content-Management-Systems durch eine zusätzliche Assistenzfunktion bei seiner Tätigkeit dahingehend zu unterstützen, dass ihm Arbeit abgenommen, Hilfestellung angeboten und Sicherheit vermittelt wird. Unter Berücksichtigung der Bedürfnisse und Fähigkeiten der Zielgruppe der Anwendung sowie der durch ausführliche Literaturrecherche gewonnenen Erkenntnisse wurde ein Assistent konzipiert, dessen für die
Interaktion relevante Benutzeroberfläche exemplarisch ausgeführt und durch empirische Untersuchung mit Probanden evaluiert wurde.

MAC Montag, 12.11.2012, 17:00 Uhr c.t., WE5/4.003

Michael Siebers: (Goal-dependent) Conditional Annulling of Actions in Automated Planning

Most modern automated planners use some kind of heuristic search to generate a suitable plan. In each search node (i.e. planning state) the successor nodes are given by the results of applying all applicable operators to the given state. So there are two major factors influencing the planning speed: the heuristic used (in calculation time and quality) and the number of applicable operators. We tackle the latter. We introduce means to expand the application preconditions of given operator definitions. Those expansions can incorporate conditions on the current state as well as conditions on the goal state (which isn't possible using the default planning problem definition language). Furthermore we show an approach to automatically learn such augmentations from previous planning experience.

Kolloquium Montag,5. 11. 2012, 14:00 Uhr , WE5/5.013

Lukas Berle (BA Präsentation, Projekt EMN-Moves): Matching natural language activity descriptions by using self-expanding ontologies -- An application to support mobility of elderly people

The aim of this work is to present and implement an approach for a self-learning system. This approach will be part of a matchmaker. The matchmaker will match elderly persons with persons who have the same interests and elderly persons with volunteers. The part of the matchmaker that is developed in this work shall assign a natural language activity description to classes in an ontology. These classes represent activities. Furthermore the approach calculates a probability from the system’s view that the natural language activity description belongs to a certain class. In addition, the approach makesuse of similarities between activities. This means that these approach is also able to return activities that are similar to the desired activities.
The approach is implemented in Java and can be tested. Some problems which negatively impact the matching quality are identified and possible solutions for these problems are presented.

Kolloquium Montag, 15.10.2012 16:00 s.t., WE5/05.005

Markus Wachter (Masterarbeit): Analysing Visual Appearance by Featural And Relational Similarity -- How Different Are Smartphones?

In this thesis the visual appearance of smartphones was analyzed by calculating a similarity value between two devices. This similarity calculation is affected by featural and relational parameters which describe the design and shape of the smartphones. A short overview of how similarity can be calculated is given and what features you can retrieve if you represent a smartphone. A possible representation of a smartphone using XML Schema is introduced. A specfic approach to similarity assessment of smart phones is proposed. Finally, a newly developped tool supporting purchase decisions for smartphones is presented.

Colloquium Archive

You find information on past talks in the Colloquium Archives [SS12]  [WS 11/12]   [SS 11]  [WS 10/11]  [SS 10]  [WS 09/10]  [SS 09]  [WS 08/09]   [SS 08]   [WS 07/08] and the colloquium archive (old)