Comparing and Contrasting Differential Evolution (DE) and Covariance Matrix Adaptation Evolution Strategies

Opawale, Samuel
Journal Title
Journal ISSN
Volume Title
University of Guelph

Evolutionary algorithms (EA) comprises population based algorithms that uses biologically inspired operators for optimization. DE and CMAES/IPOP are two powerful forms of EA that act on real numbers in order to provide solutions to multidimensional problems. Previously, researchers have tried to compare these two algorithms head-to-head, but no attempt has been made to compare and contrast the underlying mechanisms of these algorithms in order to better understand their effects and functionalities. The selection operator for CMA-ES was modified to make it more DElike by adding elitism selection instead of (μ, λ)). A new selection operator,here, was added to ES. We noticed an improvement with IPOP when here and elitism were used singly. In combination, the effect becomes remarkable, producing often a several orders of magnitude improvement in convergence time. One function, Levy, cause IPOP to stall when elitism was added. The reason for this is currently unknown.

Differential evolution, Covariance Matrix adaptation Evolution, Population based algorithms, Biologically inspired operators