Generative models for statistical mechanics

Generative adversarial networks and other generative models learn the probability distribution of a system and can both generate different systems with the same statistical properties and discriminate between systems with different statistics. In principle, they can be trained on physical systems in different local minima of their energy landscape. In the spirit of metadynamics, the simulation might encourage the exploration of a different part of the landscape if the configuration is likely known according to the discriminative network, and the generative network should have fundamental connections to the statistical mechanics of the system.

People

Jorge A. Muñoz

BS UTEP '07. MS, PhD Caltech '09, '13

Reynaldo Martinez

Physics and Computer Science