DeepMind Researchers Develop A Machine Learning Technique For Accurate Sampling And Free-Energy Estimate Of Solid Materials Using Normalizing Flows

A significant challenge of computational statistical mechanics is the accurate estimation of equilibrium parameters of a thermodynamic system. For decades, the methods of choice for sampling such systems at large have been molecular dynamics (MD) and hybrid Monte Carlo. Strategies for sampling probability distributions have increased, and most try leveraging normalizing flows.

Normalizing Flows are a technique for creating complicated distributions that involve changing a probability density through a sequence of invertible mappings. These are desirable because of 2 characteristics: …

deepmind energy free learning machine machine learning machinelearningnews materials researchers sampling solid