My research interests are in generalization to unseen inputs and tasks (robustness to distribution shifts, extrapolation), learning with unlabeled data and limited supervision (transfer learning, semi-supervised learning, unsupervised learning), and applications/motivations from real-world robustness problems and NLP.