A Knowledge Transfer Mechanism Based on Population Distribution Information for Multifactorial Differential Evolution
Ye Zou, Deming Peng, Yiqiao Cai · 2021
Evolutionary multi-task optimization (EMTO) is a novel optimization paradigm that improves the performance of each task by transferring valuable knowledge across related tasks. At present, a significant challenge and research direction in EMTO is to select useful knowledge to reduce the negative transfer phenomenon. In most EMTO algorithms, however, population distribution information of different tasks has not been effectively used in designing the mechanism for transferring useful knowledge. Based on this consideration, we propose a novel knowledge transfer mechanism based on population distribution information to alleviate the problem of negative transfer. Further, to evaluate the effectiveness of the proposed mechanism, an improved multifactorial differential evolution (MFDE) with this mechanism is presented, leading to a new EMTO variant, termed PDI-MDE. The experimental results on a suite of single-objective multitasking benchmark problems have demonstrated the competitive performance of PDI-MDE.