Evolutionary Multi-Task Optimization With Adaptive Intensity of Knowledge Transfer

Evolutionary multi-task optimization (EMTO) aims to solve multiple optimization tasks simultaneously via cross-task knowledge transfer, which has attracted considerable attention in the community of evolutionary computation. For EMTO, the intensity of knowledge transfer is one of the most crucial factors for the algorithm performance, which has a close relationship with the relatedness between tasks. In this work, to adaptively tune the intensity of knowledge transfer, a new EMTO algorithm is proposed by capturing the relatedness between tasks. First, the population is divided into two groups based on the skill factor, an index of the task where the individual performs the best across all tasks. Then, a relatedness matrix is defined by comparing the fitness values of individuals from the two groups. Based on the relatedness matrix, the relatedness among different tasks is quantified, so that the intensity of knowledge transfer can be synchronously tuned. In addition, in the proposed approach, a knowledge archive is designed to save the successfully transferred individuals for further improving transfer effectiveness. Extensive experiments on two widely recognized test suites are carried out, and 8 well-established EMTO algorithms are included in the performance comparison. The results demonstrate that the proposed approach has very competitive performance.