universal domain adaptation (UniDA) aims to transfer knowledge from the
source domain to the target domain without any prior knowledge about the label
set.
The challenge lies in how to determine whether the target samples belong to
common categories.
The mainstream methods make judgmen