Personal tools

SC²S Colloquium - April 23, 2014

From Sccswiki

Jump to: navigation, search
Date: April 23, 2014
Room: 02.07.023
Time: 3 pm, s.t.


Josef Niedermayer: Development of an Interactive Coupling for Simulation Programs

This thesis describes the coupling of two simulation programs. First an overview of the coupling of software is given. Afterwards the two programs, Athlet and Modelica, to be coupled are introduced. Modelica is a modern modeling and simulation language. Its predefined classes allow to simulate physical models form a wide range of areas. Athlet is a computer program for the simulation of accidents in the cooling circuit of nuclear reactors. The two programs are coupled the following way: The module of Athlet which is responsible for the simulation of the control system of a reactor is replaced by a model designed in Modelica. The coupling itself is done in Python language. This means Athlet and Modelica are controlled by the Python interpreter. The coupling mechanisms are applied to a specific problem. They are partially validated and its performance is analyzed.


Sebastian Soyer: Nonlinear Density Estimation with Applications in Astronomy

Density estimation is a basic task in data mining. Approaches range from kernel based techniques which are computationally expensive for large data sets to grid based methods that rather depend on the dimensionality of the processed data. We present the grid based "Maximum A Posteriori" approach which approximates the desired density function with a grid. In order to tackle the curse of dimensionality a sparse grid is applied. It makes the approach applicable for low to moderate dimensional problems. Furthermore a comparison is drawn to kernel based methods on density estimation based applications such as classification on synthetic and real world data sets.


Johann Maier: Online-Lernmethoden für Data Mining mit Dünnen Gittern

Da die heutzutage verfügbare Menge an Daten allein durch die menschliche Auffassungsgabe nicht mehr bewältigt werden kann, ist der Einsatz von leistungsstarken Werkzeugen und Techniken zur Auswertung dieser unumgänglich. Diese Arbeit beschreibt die Entwicklung eines Algorithmus, der die effiziente Behandlung hochdimensionaler Probleme des Data Mining ermöglicht. Hierfür wurde die Theorie der dünnen Gitter, sowie die des stochastischen Gradientenverfahrens untersucht und daraus der Algorithmus abgeleitet. Im Anschluss wurden Experimente auf Basis von diversen Regressionsproblemen durchgeführt, um die Funktionalität des Algorithmus zu bewerten.