Scientific Foundations Models — an initiative in the Michigan Institute for Computational Discovery and Engineering (MICDE) at the University of Michigan, launched in partnership with Deep Forest Sciences — aims to leverage large amounts of unlabelled data available in scientific domains to develop scientific foundation models enabling Generative AI for materials design and towards autonomous scientific discovery agents.
We’re training BERT-based models to accelerate material discovery and enable Generative AI for molecular design.
Creating surrogate physics models by finetuning a foundation model trained on multiple PDEs.