CS231N Spring lecture14 Reinforcement Learning Lecture Notes

(Not very clear, next time to listen again)

1. Enhance learning

There is an Agent and environment interaction. At t time, the Agent learns that the state is St, making the action is at;environment on the one hand to give reward signal RT, on the other hand change the state to st+1;agent to obtain RT and st+1. The goal is for the Agent to learn some kind of mapping of St to at π* to maximize the cumulative reward,∑γtrt, where γt is the discount factor (discount factor).

Describe the RL problem with Markov decision process. Markov process is a process with Markov properties. Markov properties: The future state depends only on the current state, or the process does not have a memory trait.

CS231N Spring lecture14 Reinforcement Learning Lecture Notes