Revealing the Inner Dynamics of Evolutionary Algorithms with Convection Selection
[ 1 ] Instytut Informatyki, Wydział Informatyki i Telekomunikacji, Politechnika Poznańska | [ P ] employee
2023
chapter in monograph / paper
english
- evolutionary algorithms
- evolutionary dynamics
- design and analysis of algorithms
- convection selection
EN Evolutionary algorithms are stochastic algorithms so they tend to find different solutions when run repeatedly. However, it is not just the solutions that vary – the very dynamics of the search that led to finding these solutions are likely to differ as well. It is especially in the algorithms with complex population structures – such as convection selection where a population is divided into subpopulations according to fitness values – where an opportunity for highly diverse dynamics arises. This work investigates the way evolutionary dynamics of subpopulations influence the performance of evolutionary algorithms with convection selection. We emploi a demanding task of evolutionary design of 3D structures to analyze the relation between the properties of the optimization task and the features of the evolutionary process. Based on this analysis, we identify the mechanisms that influence the performance of convection selection, and suggest ways to improve this selection scheme.
24.07.2023
491 - 494
20
140