Machine Learning for Atomic-Scale Modeling - Potentials and Beyond - IPAM at UCLA

Explore machine learning techniques for atomic-scale modeling in this comprehensive lecture by Michele Ceriotti from École Polytechnique Fédérale de Lausanne (EPFL). Delve into the world of interatomic potentials based on machine learning and their impact on extending the length and time scales accessible to explicit atomistic simulations. Discover solutions to challenges in the field, including poor scaling with chemical diversity and the incorporation of functional properties beyond interatomic potentials. Learn about a scheme that compresses chemical information to reduce model costs, and its application in constructing a potential for 25 d-block transition metals. Examine the convergence of quantum and data-driven approaches through examples of simulations involving electronic charge density, single-particle Hamiltonian, and electron density of states for matter with thermally-excited electrons.