AbstractsComputer Science

A meta-algorithmic procedure is a computer procedure that operates upon another algorithm and its associated design space to produce another algorithm with desirable properties (e.g., faster runtime, better solution quality, ...; see e.g., Hoos [2008]). Many meta-algorithmic procedures have runtimes that are dominated by the runtime of the algorithm being operated on. This holds in particular for automatic algorithm configurators, such as ParamILS, SMAC, and GGA, which serve to optimize the design (expressed through user settable parameters) of an algorithm under certain use cases. Consequently, one can gain improved performance of the meta-algorithm if evaluations of the algorithm under study can be done in parallel. In this thesis, we explore a distributed version of the automatic configurator, SMAC, called pSMAC, and the library, AEATK, that it was built upon, which has proved general and versatile enough to support many other meta-algorithmic procedures.