Provably Efficient Reinforcement Learning Under Linear Model Structures: From Tabular To Feature Based Exploration