Thursday 27 Oct 2016: Let the cloud design your next multi-objective optimizer
Manuel Lopez-Ibanez - University of Manchester
Harrison 170 14:30-15:30
Thanks to the vast computational power available nowadays from commodity clusters and the progress in general-purpose automatic algorithm configuration methods, it has become possible to automatically find a high-performing parameter configuration of an algorithm even in the presence of heterogeneous problem instances and hundreds of numerical and categorical parameters. When computation power and automatic configuration methods are coupled with a flexible algorithmic framework of algorithmic components, new algorithmic designs may be found automatically that outperform those proposed by human designers. Such an approach has been shown to work particularly well for multi-objective optimizers, such as ant colony optimization and evolutionary algorithms. However, the use of such approaches also raises urgent questions about quality metrics, benchmark problems, and the "novelty" of algorithms in multi-objective optimization.