Domain Adaptation with Invariant Representation Learning: What Transformations to Learn?