Exploration of truss metamaterials with graph based generative modeling

In the expanding landscape of metamaterial design, Zheng and colleagues introduces a framework that bridges design and properties, using machine learning to enhance truss metamaterials. A neural network creates an interpretable, low-dimensional space, empowering designers to tailor mechanical properties. Optimisation tasks in the inverse design of metamaterials with machine learning were limited due to the representations of generative models. Here the author comments a recent publication in Nature Communications which generates a latent space representation that unlocks non-linear optimisations.