SchNet - A Deep Learning Architecture for Molecules and Materials

Explore the groundbreaking deep learning architecture SchNet, designed specifically for modeling atomistic systems in chemical physics. Delve into the application of continuous-filter convolutional layers to accurately predict properties across chemical space for both molecules and materials. Discover how SchNet learns chemically plausible embeddings of atom types throughout the periodic table. Gain insights into the integration of kernel methods with SchNet to generate highly accurate and predictive results. This 49-minute talk by Huziel E. Sauceda at BIMSA showcases the paradigm shift in artificial intelligence within chemical physics and demonstrates the potential of deep learning in representing quantum-mechanical interactions and enhancing the exploration of chemical compound space.