I am broadly interested in interactive machine learning, including active learning, contextual bandits, and reinforcement learning. My goal is to design algorithms with both statistical and computational efficiencies. Specifically, I study the following two aspects: Sample complexity. Design algorithms that can learn good classifiers/polices with as few samples as possible, thus minimizing labeling efforts or human interventions. Runtime and memory requirement. Once statistical efficiency is achieved, design practical algorithms that have runtime and memory requirement polynomially in problem-dependent parameters. I am also interested in incorporating (deep) representation learning into the field of interactive machine learning, from both theoretical and empirical perspectives.