Performance evaluation of the SOMA asynchronously parallel distribution

Abstract

Self-Organising Migrating Algorithm (SOMA) is a very effective tool of evolutionary optimization. This paper deals with performed evaluation of asynchronous parallel version of SOMA, based on the strategy AllToOne. The experiment was focused on the dependence of the solution quality (finding the global extreme value or approximating to global extreme value) on the number of used algorithm threads. In total, 10 different test functions were employed. Each function had 100 dimensions to optimize and the optimization was always repeated 100 times for 1,2,3 and 4 threats. Overlay 4000 independent runs of SOMA were rune and statistically evaluated.