Trustworthy Representation Learning Across Domains