Measuring population diversity in variable dimension search spaces

Abstract

Measuring diversity in evolutionary algorithms presents a complex challenge, especially in optimization tasks with variable dimensionality. Current literature offers limited insights on effectively quantifying diversity under these conditions. This paper addresses this gap by evaluating the effectiveness of conventional diversity measures in variable dimension contexts and identifying their limitations. We introduce a novel diversity measurement approach tailored to these dynamic environments. Our method comprehensively captures both the structural and parametric diversity of populations, providing a more nuanced understanding of diversity changes over time. Through a series of experimental scenarios, we demonstrate that our proposed measure effectively tracks the evolution of diversity in populations with variable dimensions.