Machine Learning for Atomic Scale Systems

This lecture by the instructor provides an overview of applying machine learning to atomic scale systems. Starting with a historical introduction to atomistic modeling, the lecture delves into the transition from empirical to quantitative descriptions of matter. The instructor explains the challenges of modeling emergent physics from first principles and introduces the concept of using machine learning as a bridge to enhance accuracy in sampling atomic systems. The lecture covers the importance of symmetry in feature mapping, the construction of symmetric features for atomic structures, and the integration of rotations to create rotationally invariant descriptors. The instructor also discusses the use of linear regression models to approximate potential energy functions, highlighting the convergence of machine learning and physics-inspired approaches.