Evolutionary Generative Design: Integrating Machine Learning and Evolutionary Computation for Automated Design Space Exploration

This study proposes an innovative framework called Evolutionary Generative Design (EGD), which integrates evolutionary computation with machine learning. In conventional evolutionary computation, the mapping between the phenotype space and the genotype space (encoding and decoding) has relied heavily on the designer’s experience and expertise, posing significant challenges to automation. EGD utilizes autoencoders to automate the mapping process using existing data, enabling the dynamic generation of design spaces. EGD enables efficient exploration and evolutionary design generation by employing these generated design spaces as genotype spaces within evolutionary computation. The automatic creation of genotype spaces allows for dynamic searches of novel and optimized designs, significantly advancing the design process. To demonstrate the effectiveness of the proposed framework, this study constructs an interactive evolutionary computation (iEC) system and applies it to multiple design scenarios, showcasing improvements in customization. Variational Autoencoders (VAE) are employed as the autoencoder model in this framework. These findings highlight EGD’s potential to revolutionize design exploration, paving the way for creative and data-driven design innovation.