Jülich Parameter Exploration (JuPeX)
Summary
Carried out in collaboration with Prof. Wolfgang Maass's research group at the Technical University of Graz, this project enables large scale parameter space exploration and optimization of simulations on HPC systems. By extending the UNICORE and JUBE frameworks developed at the JSC, we will provide a flexible platform for parameter exploration using well known optimization algorithms.
Project description
Exploring the consequences of parameter changes on the observables in simulation or analytical software is a common practice and general challenge in all computational sciences, as a generalization of the experimental method to numerical analysis. We are implementing a highly-parallel framework enabling large scale parameter space exploration and optimization of simulations on HPC systems.
Short term goal
Use machine learning as a global optimizer of parameters for different simulation kernels on HPC.
Immediate technical goal
Extend UNICORE and JUBE to run an optimizing kernel and a fitness function to vary parameters for ensembles of computational kernels.
![Fig 1: Search patterns for various optimization methods [Diaz-Pier et. al. 2017].](/en/ias/jsc/images/slns/jupex/@@images/image/great)
Our contribution
Integration of machine learning algorithms with the batch systems used on the Jülich Supercomputing Centre’s supercomputers and benchmarking of the performance of long running optimization sessions.
Our collaboration partners
This project is being conducted in collaboration with the research group of Prof. Dr. Wolfgang Maass at the Graz University of Technology.
References
S. Diaz-Pier, C. Nowke, and A. Peyser. Efficiently navigating through structural connectivity parameter spaces for multiscale whole brain simulations. In 1st Human Brain Project Student Conference, page 86, 8-10 Feb. 2017.
