Machine Learning in Molecular Dynamics

This lecture explores the application of machine learning in molecular dynamics simulations, focusing on the curse of dimensionality in solving the Schrödinger equation numerically. It discusses the unreasonable effectiveness of machine learning in various tasks and the ABC of neural network representation and training. The lecture delves into the approximation power of neural networks, global convergence, dynamical Universal Approximation Theorem, and force-field estimation. It also covers free energy calculations, Boltzmann sampling using deep learning, and concludes by highlighting the potential of machine learning to revolutionize molecular dynamics simulations.