Advances in Meta Reinforcement Learning

This chapter explores the recent advances in meta reinforcement learning, including techniques like model agnostic meta learning and the lottery ticket procedure. It highlights the potential of end-to-end tuning of hyperparameters and the importance of inductive biases in adapting quickly to new tasks. The chapter also discusses the long-term goals and potential of meta RL in creating systems that self-referentially refine themselves in a hierarchical loop.