The Role of Embedding Complexity in Domain-invariant Representations

Abstract

Unsupervised domain adaptation aims to generalize the hypothesis trained in a source domain to an unlabeled target domain. One popular approach to this problem is to learn a domain-invariant representation for both domains. In this work, we study, theoretically and empirically, the explicit effect of the embedding on generalization to the target domain. In particular, the complexity of the class of embeddings affects an upper bound on the target domain's risk. This is reflected in our experiments, too.

Tasks

Reproductions