Adaptive Multitask Evolutionary Optimization for Tasks with Non-Uniform Evaluation Cost

Recently, there has been a great deal of research on evolutionary multitasking. Most existing evolutionary multi task algorithms treat all optimization tasks equally, assuming that all tasks can be evaluated at the same time. However, in practice, the function evaluation time of different tasks varies greatly due to their distinct properties. This paper proposes to formalize this special scenario of multi task optimization as multi task optimization with non-uniform evaluation cost, and concentrates on how to effectively and efficiently solve it. For this type of problem, we propose an adaptive evolutionary multi task optimization algorithm that uses a multi-population evolutionary framework to solve multiple tasks in parallel, allocating different computational resources to different tasks. It performs effective positive knowledge transfer between different tasks, and dynamically adjusts the amount of knowledge transfer according to the survival rate of offspring and transferred individuals. To verify the performance of the proposed algorithm, we conduct experiments on artificially constructed benchmark problems fitting our proposed scenario of multi task optimization with non-uniform evaluation cost. The experimental results show the superiority of our proposed algorithm over other state-of-the-art algorithms. For the multitask optimization problem with non-uniform evaluation cost, the proposed algorithm can solve tasks efficiently within a limited budget of computational resources.