Learning trivializing gradient flows for lattice gauge theories

We propose a unifying approach that starts from the perturbative construction of trivializing maps by L\"uscher and then improves on it learning. The resulting continuous normalizing flow model can be implemented using common tools lattice field theory requires several orders magnitude fewer parameters than any existing machine learning approach. Specifically, our achieve competitive performance with as few 14 while deep-learning models have around 1 million for $SU(3)$ Yang--Mills $16^2$ lattice. This has obvious consequences training speed interpretability. It also provides plausible path scaling machine-learning approaches toward realistic theories.