My main research interest is in natural language processing. My goal is to equip intelligent agents with reasoning capacities and make them reliable and robust in real scenarios (e.g., medical). Some directions I'm excited nowadays include:
(1) Question Answering: How to use unstructured text and structured tables as knowledge to answer questions that require complex reasoning (e.g., multi-hop QA);
(2) Controllable Text Generation: How to generate text (e.g., provide long-form answer to a question) that satisfies interlocking constraints: being controllable and truthful.
(3) Model Uncertainty and Calibration: how to ensure agents provide reliable confidence estimates present to human in an understandable way.