State of the Art on domain shifts characterization between source annotated data and new target data

A fundamental hypothesis running through the field is that bringing closer the two domains in the sens of a certain distance function facilitates the understanding of the target domain by the model, and is therefore beneficial for its performance in the target domain. We focus on the verification of this hypothesis in a simplified context. This is a work-in-progress, and in the present document we present our line of attack as well as certain theoretical considerations on the distance functions that are widely used in UDA literature.