rl_review

Reinforcement learning review collection §

tag:#rl#review

A Tour of Reinforcement Learning: The View from Continuous Control §

Benjamin Recht, Arxiv 2018

• Define reinforcement learning
  • Contemporary RL
    • Episodic reinforcement learning: want to find a policy that maximizes the reward with the fewest total number of samples computed by the oracle. From experienced trajectories and reward to compute a new policy.
  • connection to supervised learning
    • supervised learning is just for prediction.
    • reinforcement learning generate u (action for control) to make reward large.
  • Strategies for solving reinforcement learning problems
    • Model-base
    • Model-free