Abstract: |
In this paper we focus on an important source of problem-difficultyin (online) dynamic optimization problems that has so far receivedsignificantly less attention than the traditional shifting of optima.Intuitively put, decisions taken now (i.e. setting the problemvariables to certain values) may influence the score that can beobtained in the future. We indicate how such time-linkagecan deceive an optimizer and cause it to find a suboptimal solutiontrajectory. We then propose a means to address time-linkage:predict the future by learning from the past. We formalize thismeans in an algorithmic framework. Also, we indicate why evolutionaryalgorithms are specifically of interest in this framework. We haveperformed experiments with two new benchmark problems that containtime-linkage. The results show, as a proof of principle,that in the presence of time-linkage EAs based upon this frameworkcan obtain better results than classic EAs that do not predict the future. |
