Explore a seminar on weak convergence and invertible coarse-graining in probability theory and molecular dynamics. Delve into Prof. Grant M. Rotskoff's discussion of combining force-matching based coarse-graining with invertible neural networks to invert coarse-graining maps statistically. Learn about the thermodynamic equivalence principle, challenges in coarse-graining biomolecular systems, and the application of machine learning in this context. Discover how this approach can recover fine-grained free energy surfaces from coarse-grained sampling in non-trivial biomolecular systems. Gain insights into metastability in protein folding/misfolding and the capture of large- and small-scale observables through this innovative method.
Overview
Syllabus
Intro
The necessity of coarse-graining
Challenges in coarse-graining biomolecular systems
Why use machine learning in this setting?
Invertible, state-dependent coarse-graining
Flexible and useful embeddings
Thermodynamic consistency (Noid and Voth)
Weak thermodynamic consistency
Rigorously inverting the CG sampling
Necessary Sacrifices
Metastability in protein folding / misfolding
Large-and small-scale observables well-captured
Taught by
Cambridge Materials
