Analysing knowledge transfer in SHADE via complex network

Abstract

In this research paper a hybridization of two computational intelligence fields, which are evolutionary computation techniques and complex networks (CNs), is presented. During the optimization run of the success-history based adaptive differential evolution (SHADE) a CN is built and its feature, node degree centrality, is extracted for each node. Nodes represent here the individual solutions from the SHADE population. Edges in the network mirror the knowledge transfer between individuals in SHADE's population, and therefore, the node degree centrality can be used to measure knowledge transfer capabilities of each individual. The correlation between individual's quality and its knowledge transfer capability is recorded and analyzed on the CEC2015 benchmark set in three different dimensionality settings-10D, 30D and 50D. Results of the analysis are discussed, and possible directions for future research are suggested.