A study of similarity measure between tasks for multifactorial evolutionary algorithm

In contrast to the traditional single-task evolutionary algorithms, multi-factorial evolutionary algorithm (MFEA) has been proposed recently to conduct evolutionary search on multiple tasks simultaneously. It aims to improve convergence characteristics of the tasks to be tackled by seamlessly transferring knowledge among them. Towards superior multitasking performance, the evaluation of task relationship plays an important role for grouping the related tasks, and solve them at the same time. However, in the literature, only a little work has been conducted to provide deeper insights in the measure of task relationship in MFEA. In this paper, we thus present a study of similarity measure between tasks for MFEA from three different perspectives. 21 multitasking problem sets are developed to investigate and analyze the effectiveness of the three similarity measures with MFEA for evolutionary multitasking.