Tractably modelling distributions over manifolds has long been an important
goal in the natural sciences. Recent work has focused on developing general
machine learning models to learn such distributions. However, for many
applications these distributions must respect manifold symmetri