This package contains utilities for the implementation of normalising flows as a generative model using Pytorch.
Release Notes
This package contains utilities for the implementation of the method of normalising flows as a generative model for lattice field theory. The method of normalising flows is a powerful approach in generative modelling that aims to learn complex probability distributions by transforming samples from a simple distribution through a series of invertible transformations. It has found applications in various domains, including generative image modelling. Normflow currently supports scalar theories in any dimension, and we are extending the package to accommodate gauge theories.
Future Plans
We plan to extend our work to the more physically relevant SU(2) and SU(3) gauge theories in four spacetime dimensions. We need to construct and train a neural network to map the link variables of a gauge theory to a theory with decoupled link variables. However, a naive transformation of the link variables may lead to a distribution that is not symmetric under gauge transformation. To avoid this problem one should construct a neural network that respects the gauge symmetry. We are currently working on the construction of neural networks that encode these gauge-equivariant transformations.