My research focuses on lifelong learning and deep sequential models. For one thing, we would like to build models that learn in a never-ending fashion. This is in stark contrast with traditional train/test pipeline of machine learning; therefore, we are investigating what kind of representations would enable lifelong learning. For another, we are working on continuous-time models for learning sequential data, e.g., physical systems, videos, handwriting, etc. We are particularly interested in neural ordinary differential equations for approximating dynamical systems, and also diffusion models that are capable of fitting temporal datasets.