abstract
- In recent years, evolutionary computation has signif-icantly advanced in processes related to machine learning. How-ever, the reciprocal integration of machine learning techniques into evolutionary computation remains relatively unexplored. Machine learning can substantially enhance the understanding of processes within Multi-Objective Evolutionary Algorithms (MOEAs) by harnessing its proficiency in identifying patterns and employing data-driven approaches. Existing studies lack a comprehensive understanding of the intricate interaction between machine learning models and evolutionary algorithms, necessi-tating prioritized investigation to ensure the efficacy, reliabil-ity, and compatibility of integrated models within optimization frameworks. This paper addresses this gap by examining the behavior of using Generative Artificial Intelligence (AI) models as a population variation operator in Multi-objective Optimization Problems. Our experimental results reveal that Generative AI, particularly Distributional Adversarial Networks (DANs), sur-passes the performance of a traditional Generative Adversarial Network. Furthermore, DANs improve the population by gen-erating novel non-dominated solutions and augmenting overall performance and diversity. This study reveals the potential of the integration of Generative AI in evolutionary computation, presenting a pathway for advancements in addressing common challenges within multi-objective optimization problems. © 2024 IEEE.