Machine Learning in Molecular Dynamics

This lecture presents examples of research on the application of machine learning in molecular dynamics, materials, and chemistry. Starting with linear regression and polynomial regression, it progresses to kernel regression and neural networks. The focus is on creating meaningful features that capture relevant information and symmetries in the system. The lecture discusses a paper on using neural networks to model potential energy surfaces efficiently, emphasizing the importance of generalizability. It then delves into another paper that unifies the modeling of materials and molecules using atomic neighbor densities and Gaussian functions. The lecture concludes with a discussion on kernel regression and structured kernels in relation to molecular energy prediction.