Effects of Elitism and Population Climbing on Multiobjective MNK-Landscapes

Abstract	

Epistasis and NK-Landscapes in the context of multiobjective evolutionary 
algorithms (MOEAs) are almost unexplored subjects. We have presented 
an extension of Kauffman's NK-Landscapes to multiobjective MNK-Landscapes 
and gave some insights into their properties from a multiobjective standpoint. 
These properties allow us to meaningfully use MNK-Landscapes as a benchmark 
tool and as a means to understand better the working principles of MOEAs. In 
this work we present four multiobjective random bit climbers (moRBCs) and use 
them to study the effects or elitism and population climbing on scalable random 
epistatic problems. Each moRBC implements a different kind of elitism in order to 
understand better its working principles. We conduct experiments on MNK-Landscapes 
with M = {2, 3, 5} objectives, N = 100 bits, varying the epistatic interactions K from 
0 to 50. Results by an elitist non-dominated sorting multiobjective genetic algorithm 
(NSGA-II) are also included for comparison.