BBOB: Nelder-Mead with resize and halfruns. Using the BBOB template, we investigate how the Nelder-Mead simplex algorithm can be combined with evolutionary ideas to give a competitive hybrid approach to optimize continuous functions. We significantly improve the performance of the algorithm in higher dimension by the addition of a reshaping step of the search, to correct for a known problem in the simplex search behaviour. We also give a reasonably good population-based approach in which only a third of the individuals is fully matured, with a bias towards fitter individuals, via a variant of the Nelder-Mead method.

Keywords for this software

Anything in here will be replaced on browsers that support the canvas element