Effects of Local Mating in Inter-task Crossover on the Performance of Decomposition-based Evolutionary Multiobjective Multitask optimization Algorithms

Abstract

Recently, Evolutionary Multiobjective Multitask optimization (EMMO) was proposed as a new research topic in the field of Evolutionary Multiobjective optimization (EMO). In contrast to conventional EMO algorithms, EMMO algorithms solve multiple multiobjective optimization problems (multiple tasks) in their single run. Most EMMO algorithms have the same number of populations as the number of tasks to be solved simultaneously, and each population corresponds to a different task. The main feature of EMMO algorithms is that offspring solutions are generated by not only intra-task crossover but also inter-task crossover. Local mating in intra-task crossover improves the search performance of EMO algorithms that use uniformly distributed weight vectors during a search, such as MOEA/D. Therefore, local mating in inter-task crossover is a promising idea for EMMO algorithms. In this paper, we propose a simple extension of MOEA/D for EMMO algorithms and a local mating method in inter-task crossover based on uniformly distributed weight vectors. Through computational experiments, we examine the effects of local mating in inter-task crossover on the search performance of the proposed algorithm. Experimental results show that the local mating improves the search performance of the proposed algorithm.