Materials Discovery Using Deep Learning and Differentiable Physics

Materials informatics datasets and computational resources are growing rapidly. While machine learning offers a promising set of generic tools for extracting insight from large volumes of data, these tools are not inherently good at generalizing to new data. For this reason, tools that enable a seamless integration of physically informed priors with machine learning are crucial for building data-driven models that generalize in a useful way. I will talk about our recent work in this direction where we combine differentiable atomistic simulations with graph neural networks and meta-learning, with applications to materials modelling, optimization, and discovery. <b> </b>