Multi-criteria differential evolution: treating multitask optimization as multi-criteria optimization

Evolutionary multitask optimization (EMTO) has shown great potential and attracted increasing attention for solving multitask optimization problem (MTOP). However, most existing EMTOs still treat the internal tasks just as different related problems, rather than the component problems of the entire MTOP, which is inefficient. Therefore, this paper proposes to treat the entire MTOP as a multi-criteria optimization problem (MCOP), so as to solve the MTOP more efficiently. To be specific, the fitness function of each task in the MTOP is treated as an evaluation criterion in the corresponding MCOP. During the evolutionary process, a round-robin multi-criteria strategy is adopted to better utilize the multiple criteria. This way, we can use different evaluation criteria in different generations or stages to guide the environmental selection and population evolution in ECs, so as to find out the optimal solutions for the criteria of different tasks. Based on the above, a multi-criteria differential evolution algorithm is developed for solving MTOPs. Experiments on widely-used MTOP benchmarks and comparisons with some state-of-the-art algorithms have verified the great effectiveness and efficiency of the proposed algorithm. Therefore, treating MTOPs as MCOPs is a potential and promising direction for solving MTOPs.