SC²S Colloquium - Aug 16, 2018

From Sccswiki
Jump to navigation Jump to search
Date: Aug 16, 2018
Room: 02.07.023
Time: 15:00 - 16:00

Dominik Fuchsgruber: Integration of the SGDE-based Classification into the SG++ Datamining Pipeline

This is a Bachelor's Thesis submission talk advised by Kilian Röhner

This thesis describes the integration of the sparse grid density estimation-based classification into the datamining pipeline of the SG++ toolbox. As the sparse grid density estimation can be reduced to solving a system of linear equations, a database to store manage precomputed system matrix factorizations for different common scenarios was introduced. Furthermore, existing algorithmic modules were refactored, mainly by reorganizing the ownership of model state instances, such as the underlying grid objects and corresponding surplus vectors. Models for density estimation as well as classification were implemented as standalone tasks, therefore being accessible to end-users. The new concepts will be evaluated and compared to their previous implementation, showing that computational as well as memory complexity is maintained.

Keywords: SG++

Daniel Langer: Efficient Verlet List Implementation on KNL Processors

This is a Bachelor's Thesis submission talk advised by Steffen Seckler

Molecular Dynamics Simulations deal with the interaction of an extraordinary high amount of particles in a domain system, so efficient approaches and algorithms are necessary. One of these approaches are Verlet Lists. For this Bachelor's Thesis, an existing MD Simulation was extended with Verlet Lists and tested on the new Intel Xeon Phi (KNL) cluster of the LRZ.

Keywords: Molecular Dynamics, KNL, ls1-mardyn