Anisotropic Machine Learning Representations for Coarse-Graining

This seminar presents Arthur Lin from the University of Wisconsin-Madison discussing anisotropic machine learning representations for coarse-graining in atomistic simulations. Explore how the innovative AniSOAP (Anisotropic Smooth Overlap of Atomic Positions) descriptor addresses limitations of traditional atom-centered representations when modeling large macromolecular systems. Learn about this anisotropic generalization that effectively describes complex molecular geometries and captures orientation-dependent interactions between groups of atoms. The talk examines three case studies demonstrating AniSOAP's applications, from unsupervised analyses of liquid crystals to learning benzene energetics, showing how molecular geometry influences phase formations and configuration energetics. Discover how AniSOAP can be incorporated into generalized coarse-grained simulation frameworks and its potential for quantifying information loss in coarse-graining processes. The 20-minute Lennard-Jones Centre discussion group seminar was held on October 21, 2024.