Abstract: Most existing unsupervised domain adaptation methods rely on explicitly or implicitly aligning the features of source and target domains to construct a domain-invariant space, often using ...