Effect of Solution Information Sharing between Tasks on the Search Ability of Evolutionary Multiobjective Multitasking Algorithms

A multitask multiobjective optimization problem (MTMOP) has multiple multiobjective problems to be solved simultaneously (i.e., multitasking of multiple multiobjective problems). One approach to such an MTMOP is evolutionary multiobjective multitasking (EMOMT). EMOMT algorithms solve multiple tasks simultaneously in a cooperative manner. They are evolutionary algorithms with multiple sub-populations. Each sub-population corresponds to a single task. During the evolutionary search, the information on each solution is shared with the sub-populations. Some EMOMT algorithms have been developed by focusing on solution information sharing. However, the effect of sharing the solution information on the search ability of EMOMT algorithms is not well examined yet. Through the examination of this effect, it is expected that better EMOMT algorithms than existing ones can be developed. In our previous study, we proposed a framework of island model-based evolutionary single-objective multitasking algorithms. Using our island model, we can analyze the effect of solution information sharing. In this paper, as an extension of our previous study, we develop an island model-based EMOMT algorithm framework for MTMOPs. Through computational experiments under various parameter settings, we examine the effect of solution information sharing on the search ability of our island model.