Evolutionary Multi-task Optimization Introducing Task Assignment Based on a Search Degree

Abstract

Multi-task optimization problems are to find solutions of multiple optimization problems simultaneously. Conventional evolutionary methods for the problems maintain a population for each task, and some offspring are generated by combining individuals in different tasks. The generated offspring are randomly assigned to tasks. However, the offspring may be assigned to inappropriate tasks. We propose a method that assigns them to appropriate tasks using a new criterion called search degree. The search degree represents how fast the solution search in each task progresses. The proposed method increases the probability of assigning the offspring to a task with a small search degree. Experimental results show the proposed method is superior.