Generative Modeling and Smarter Sampling for Lattice Gauge Theories
DescriptionIn this work we describe how recent advancements in generative modeling have contributed to simulations in lattice gauge theory, and discuss some ongoing work in this direction. In particular, we are interested in generating independent (lattice) gauge configurations, distributed according to the density of our theory. In order to reliably predict values which can be experimentally measured, simulations are carried out at increasing spatial resolution and extrapolated to the continuum (infinite resolution) limit. In this limit, the cost of generating configurations (using existing techniques) is known to scale exponentially, quickly becoming prohibitively expensive, preventing further exploration. In this work we present a new technique that uses a generalized version of the Hamiltonian Monte Carlo algorithm, parameterized by weights in a neural network, that can be trained to make these simulations more efficient, thereby decreasing their overall computational cost.
TimeMonday, June 2616:30 - 17:00 CEST
Computer Science, Machine Learning, and Applied Mathematics