[PAST EVENT] Physics Seminar - Kimmy Cushman
Access & Features
- Open to the public
Speaker: Kimmy Cushman, Yale University , Title: “Replacing Markov Chain Monte Carlo with Generative Flow Neural Networks”
Quantum chromodynamics and other strongly coupled gauge theories are only solvable numerically, and the current state-of-the-art methods are variants of Markov Chain Monte Carlo (MCMC) integration over particle fields defined on a discretized spacetime lattice. Properly sampling from the underlying distribution of lattice configurations is essential to computing correct observables, but traditional MCMC computational limitations place severe constraints on the resolution of simulations which can be performed. In this talk, I will discuss progress made toward replacing Markov chain Monte Carlo approaches with generative flow neural networks for the generation of gauge configurations. I will explain the benefits of using this architecture of neural network, and show our progress in implementing a one-dimensional spin theory. In this toy model, we implement a novel approach to confirming the physical accuracy of the ensembles by using a renormalization group Monte Carlo method to verify that the configurations are in the same universality class as those produced by MCMC.