Evolutionary Transfer Optimization - A New Frontier in Evolutionary Computation Research
Kay Chen Tan,Liang Feng,Min Jiang
DOI: https://doi.org/10.1109/mci.2020.3039066
IF: 9.809
2021-02-01
IEEE Computational Intelligence Magazine
Abstract:The evolutionary algorithm (EA) is a nature-inspired population-based search method that works on Darwinian principles of natural selection. Due to its strong search capability and simplicity of implementation, EA has been successfully applied to solve many complex optimization problems, which cannot be easily solved by traditional exact mathematical approaches, such as linear programming, quadratic programming, and convex optimization. Despite its great success, it is worth noting that traditional EA solvers start the search from scratch by assuming zero prior knowledge about the task at hand. However, as problems seldom exist in isolation, solving one problem may yield useful information for solving other related problems. There has been growing interest in conducting research on evolutionary transfer optimization (ETO) in recent years: a paradigm that integrates EA solvers with knowledge learning and transfer across related domains to achieve better optimization efficiency and performance. This paper provides an overview of existing works of ETO based on the type of problems being solved by these methods, which are ETO for Optimization in Uncertain Environment, ETO for Multitask Optimization, ETO for Complex Optimization, ETO for Multi/Many-Objective Optimization, and ETO for Machine Learning Applications. The paper also highlights some of the challenges faced in this emerging research field of computational intelligence and discusses some promising future research directions in ETO. It is hoped that the study presented in this paper can help to inspire the development of more advanced ETO methods and applications.
computer science, artificial intelligence