Matrix-Driven Adaptive Dual-Space Evolutionary Algorithm for Many-Task Optimization

Evolutionary multi-task optimization is an emerging research topic in the field of evolutionary computation. It aims to achieve simultaneous optimization of different tasks by dynamically leveraging the synergies that exist between them. As the number of tasks being optimized simultaneously increases, the differences between the tasks can vary significantly, and their iterative trends may also diverge. These factors contribute to a higher likelihood of negative transfer during the knowledge transfer process. To address these challenges, this paper proposes a matrix-driven adaptive dual-space evolutionary optimization algorithm, referred to as MaKAM. The MaKAM algorithm aims to compute the similarity between tasks using dual measurement criteria from both the objective space and decision space, effectively avoiding negative transfer between tasks. For many-task single-objective optimization problems, comparative algorithms are evaluated using the CEC19MaTSO and WCCI20MaTSO test suites. For many-task multi-objective optimization problems, comparisons are made using the CEC19MaTMO and WCCI20MaTMO test suites. The results indicate that the overall performance of the proposed MaKAM algorithm significantly surpasses that of the compared algorithms in the literature.