Effective Computational Resource Allocation in Evolutionary Multi-Objective Multi-Task Optimization

Evolutionary multitasking optimization achieves efficient solutions to multi-task optimization problems through the transfer and reuse of genetic information across tasks. However, in scenarios with limited computational resources, how to allocate resources among tasks effectively remains a challenge. This paper addresses the multi-objective multi-task optimization problem and proposes a computational resource allocation strategy based on the analytics of evolutionary changes in the objective space and the unified search space. The algorithm decomposes each multi-objective task into single-objective optimization subproblems using a decomposition strategy, and constructs a dual-utility function based on the scalarizing function of the subproblems and the diversity changes in the unified search space to determine the optimization priorities of the subproblems. The experimental results demonstrate that our proposed algorithm can effectively allocate computational resources dynamically and significantly outperforms many comparison algorithms in terms of inverted generational distance related metrics.