Evolutionary multi-task optimization with hybrid knowledge transfer strategy

As an emerging research paradigm in the field of evolutionary computation, evolutionary multi-task optimization (EMTO) has received an increasing amount of attention due to its capability in concurrently solving multiple related optimization tasks. In EMTO, without any prior knowledge about the complementarity between different tasks, the task relatedness is mainly captured dynamically through the evolving population. However, how to transfer the knowledge across tasks in accordance with different degrees of relatedness as the search proceeds has received little visibility. To address this issue and further enhance the performance of EMTOs, this paper proposes a hybrid knowledge transfer (HKT) strategy. In HKT, a population distribution-based measurement (PDM) technique is designed to evaluate the task relatedness based on the distribution characteristics of the evolving population, and then a multi-knowledge transfer (MKT) mechanism is employed to conduct multiple strategies of knowledge transfer according to the degree of relatedness between tasks. By incorporating the HKT strategy into EMTO, the resultant algorithmic framework, termed EMTO-HKT, is presented. The experimental results on the single-objective multi-task optimization test suite demonstrate the superiority of EMTO- HKT compared with other state-of-the-art EMTO algorithms. (c) 2021 Elsevier Inc. All rights reserved.

Automated summary

Evolutionary Multi-Task Optimization (EMTO) is a promising research paradigm for solving multiple related optimization tasks concurrently. This paper proposes a hybrid knowledge transfer (HKT) strategy to enhance EMTO's performance by evaluating task relatedness based on population distribution and transferring knowledge according to the degree of relatedness. Experimental results demonstrate the superiority of EMTO-HKT over other state-of-the-art EMTO algorithms.