Diversity Measurement

Abstract

Chapter 1 identified previous EA research in dynamic fitness landscapes where the researchers applied several diversity-increasing techniques, largely following the intuition that a mostly converged population needs to increase its explorative capability to identify a moved optimum. Chapter 2 identified how diversity improves EA performance in dynamic fitness landscapes. Chapter 3 identified examples from biology and engineering where diversity plays a key role in providing satisfactory solutions to dynamic problems. Having established the importance of diversity to the operation of an EA in a dynamic environment, this chapter will address the measurement of population diversity. One of the problems with diversity measurement is that, historically, it has been computationally expensive. The first section of this chapter will address historical methods of measuring diversity, and introduce a mathematical innovation that provides an efficient method for computing the most common population diversity measures. The second section of this chapter will address the shortcomings of the historical measures of diversity as applied to EAs in dynamic fitness landscapes, and will extend the techniques developed in the first section to derive and present a more useful measure of population diversity for dynamic environments called the “dispersion index.”