Abstract
In this presentation, we will discuss opportunities for accelerating the simulation of physical systems at equilibrium with machine learning. These methods rely heavily on deep generative models, which are highly flexible probabilistic models capable of providing independent samples of complex distributions at negligible cost. They can be used to facilitate the simulation of a Boltzmann distribution
Boltzmann distribution, a task generally difficult either because of dimensionality, multi-modality, poor conditioning or a combination of the above. We will present different methods combining classical simulation and learning, some application examples and discuss the remaining challenges.