My research focuses on improving deep learning for practical problems to better serve the end users, such as learning on data with distributional bias, privacy concerns, and scarce labels. I have been working on federated learning, personalization, synthetic data, gradual domain adaption, meta-learning, and deep imbalanced learning.