Finding interesting things

Model- and simulation-designers are often interested not in the optimum output of their system, but in understanding how the output is sensitive to different parameters. This can require an inefficient sweep of a multidimensional parameter space, with many samples tested in regions of the space where the output is essentially all the same, or a sparse sweep which misses crucial “interesting” regions where the output is strongly sensitive. In this paper we introduce a novel population-oriented approach to adaptive parameter sweeping which focuses its samples on these sensitive areas. The method is easy to implement and model-free, and does not require any previous knowledge about the space. In a weakened form the method can operate in non-metric spaces such as the space of genetic program trees. We demonstrate the method on three test problems, showing that it identifies regions of the space where the slope of the output is highest, and concentrates samples on those regions. Cate...

• Type: Conference
• Year: 2007
• Where: GECCO
• Authors: Sean Luke, Deepankar Sharma, Gabriel Catalin Balan
• Keywords: GECCO 2007, Inefficient Sweep, Multidimensional Parameter Space, Optimum Output