Kedar Hippalgaonkar1,2
Nanyang Technological University1,Institute of Materials Research and Engineering2
Kedar Hippalgaonkar1,2
Nanyang Technological University1,Institute of Materials Research and Engineering2
We present a generative model framework for rapidly discovering inorganic materials. By leveraging deep generative models and a property prediction target learning branch, the model efficiently explores the chemical space and identifies materials with desirable properties (in this case, formation energy and synthesizability[1]). The generative model consists of constructed latent space representation from known materials and generating new materials by sampling from this space. The main ingredient is our newly designed reconstructable Wyckoff site based crystal representation that includes symmetry considerations.<br/><br/>Beyond reducing reconstruction loss, to validate the generated materials, Density Functional Theory has been performed in addition to high-throughput synthesis and validation of the generated compounds.<br/><br/>In summary, our generative model framework offers a promising approach for accelerated inorganic materials discovery. Its integration with robotic, high-throughput workflows and self-driving labs ensures efficient validation of the generated materials, facilitating their successful translation into real-world applications.<br/><br/>[1] R. Zhu et. al. 2023 ACS omega 8 (9), 8210-8218)