Knowledge transfer in evolutionary multi-task optimization: A survey

Ziying Tan, Linbo Luo, Jinghui Zhong

Abstract: Highlights•Evolutionary multi-task optimization (EMTO) is an emerging paradigm of evolutionary computation. This paper presents a comprehensive survey of the EMTO algorithms with a specific focus on their knowledge transfer (KT) methods, which are of critical importance to the success of EMTO.•A multi-level taxonomy is proposed that decomposes the design of KT into the key stages involved, major approaches for each stage, and various strategies for realizing different approaches. The taxonomy facilitates a systematic and in-depth analysis of how KT is exercised in EMTO.•The possibility and difficulty level of integrating various methods at the two key stages (i.e., when to transfer and how to transfer) of KT are assessed, which is helpful for practitioners to devise more advanced algorithms to improve the performance of KT at both the two key stages.•The feasibility of applying the transfer learning approaches to various categories of EMTO algorithms according to our taxonomy is evaluated, which is helpful to expand the readers’ understanding of how to leverage transfer learning techniques to the design of EMTO.•Future directions for improving knowledge transfer performance in EMTO are discussed.