A Survey on Evolutionary Multitask Optimization

Abstract

Traditional evolutionary algorithms (EAs) typically search from scratch. In the modern era of artificial intelligence(AI), solving multiple, related problems simultaneously has become a critical challenge. In practical applications, people often need to handle and complete multiple tasks simultaneously. Based on this demand, evolutionary multitask optimization (EMTO) has emerged as a powerful paradigm within the broader field of computational intelligence. EMTO is considered an effective method to provide optimal solutions for each specific task by promoting knowledge transfer between different optimization tasks, mirroring concepts like transfer learning and multitask learning in mainstream AI. Due to the powerful ability of EMTO in parallel search, it has attracted significant attention from researchers in the field of evolutionary computing, thus promoting extensive research on the application of EMTO. This paper comprehensively reviews the progress of EMTO research in recent years using a manual and systematic literature review methodology. First, we provide an in-depth mathematical description of the evolutionary multitask optimization problem and introduce the core framework of the multi-factorial EA. Subsequently, the basic implementation methods of EMTO and its various extensions are classified and summarized. Finally, we discuss in detail the theoretical analysis, benchmark problems, and practical applications of EMTO. At the end of this paper, we summarize the content and propose possible directions for future research.