At [AMD] our research focus is on developing and using predictive methods for the discovery of molecules and materials for energy applications. Our current research activities include: 1) Physics-based computing, 2) Molecule/material databases, 3) Discriminative models, 4) Generative models, and 5) Data visualization.
1) Physics-based computing: We develop and use high-throughput computational screening workflows for the classical and quantum chemical modelling of compounds at the nanoscale. This way we generate trustworthy computational (i.e. synthetic) data on chemical compounds, which is an add-on to existing molecule and material databases.
2) Molecule/material databases: FAIR data is the foundation of AI. To improve AI-aided design of molecules and materials, we locally generate or collect, merge, organize, and label the existing datasets in a unifying manner towards a generalizable approach.
3) Discriminative models: For the forward design of molecules and materials, we use various data-driven AI models and their hybrid forms to draw reliable predictions of properties directly from line notation representation of molecules and materials. We integrate useful data into refined models and ontologies and execute them in low-data environments, thereby improving the accuracy of predictions.
4) Generative models: For the reverse design of materials, we use deep generative models with an aim to enlarge the applicability domains of the models. This way we can design completely new molecules and materials that have not been considered before for the target energy storage or conversion application.
5) Data Visualization: We develop software tools to handle high-dimensional data in smart ways, such as by using modern dimensionality reduction techniques and similarity algorithms, which are an integral piece of diversity analysis, outlier detection, virtual chemical library design, screening, and the follow-up engineering of compounds.