The system shall autonomously plan suitable sequences of performance experiments, execute them, and analyze their results. The knowledge on algorithm performance obtained this way shall then be incorporated into existing algorithm selection mappings. To automate this particular kind of performance data post-processing, incremental methods for learning algorithm selection mappings shall be investigated. The learned selection mappings shall be integrated into the simulation system James II, in a manner that allows fully automatic re-configuration depending on the models to be simulated and also on the available hardware. Both aspects of the project, empirical performance analysis and incremental algorithm selection, complement each other: the first aspect is required to gather sufficiently many data, while the second aspect analyzes these data to improve the overall performance of the simulation system.
Further information on this research project in the WWW: