Bachelor's thesis presentation. Frederik is advised by Atamert Rahma, Samuel Newcome, and Prof. Dr. Felix Dietrich.
Previous talks at the SCCS Colloquium
Frederik Willemsen: Data-Driven Selection of Algorithmic Configurations in Particle Simulations
SCCS Colloquium |
In Molecular Dynamics simulations, it is important to analyze characteristics of the particles. Modern simulation tools like AutoPas provide auto-selection of different algorithmic configurations at runtime, which can be time-consuming when there are many configurations to choose from. This thesis aims to implement a new tuning strategy called Cluster-Based Tuning, which groups simulation data into clusters and assigns configurations to each cluster that are expected to perform well. We observed speedups between 4 to 28 times in the tuning phases in two different scenarios against full-search.