Offline Learning
Research
My academic research falls into two main areas: Reinforcement learning research and applications, where I develop novel algorithms to handle well-known difficult problems in RL training, such as offline learning, uncertainty-aware RL, semi-supervised RL to handle data limitation and sub-optimality problems.
My other main research agenda uses advanced RL methods combining with uncertainty theory to help predict user’s preference on movies (recommender system, multi-modal fusion), predict ASD patients from time series data recording from their reactions when playing a game (medicine, time sequence prediction), and CV related image based sketch retrieval, object detection, etc.
Now, my focus is on applying RL to language models for text summarization and traditional backbone networks for architecture search to solve Lottery Ticket Hypothesis problem.
