Domain adaptation (DA) learning occurs often in practice. It refers to situations in which the training data available to the learner is (generated by a distribution that is) different than (the distribution generating) the target task that the learnt predictor will be evaluated on. Theoretical work on domain adaptation aims to characterize situations in which such learning is possible and to propose appropriate learning paradigms and algorithms. In this talk I will survey the current theoretical understanding of DA learning and list some directions in which further research may require collaboration with people that encounter such learning issues in practice.