Visualizing Population Dynamics to Examine Algorithm Performance

This work assesses the efficacy of evolutionary algorithms (EAs) using an intuitive multidimensional scaling (MDS) visualization evolution a population. We propose use landmark MDS (LMDS) to overcome computational challenges inherent visualizing many-objective and complex problems with MDS. For benchmark we tested, LMDS is akin visually, whilst requiring less than 1% time memory necessary produce same objective space solutions, leading possibility online visualizations for multi- optimization evaluation. Using from DTLZ WFG test suites, analyze how Landmark can offer far greater insight into algorithm performance traditional metrics such as hypervolume alone, be used complement explicit metrics. Ultimately, this allows visual identification problem features assists decision maker in making recommendations parameters/operators creating testing better EAs solve problems.