RL reinforcement learning summary (2)

Markov Decision Process

Markov decision process, English is Markov Decision Process, referred to as MDP.

Requirements for Markov Decisions

1. Can reach the ideal state.It can also be said that it can reach the final state.For example: AlphaGo can play chess to the winning step
2. Various attempts can be made
For example: AlphaGo can choose one of multiple positions on the chessboard when making a move
3. The next state of the agent is only related to the current state and the actions taken in the current state, not related to the previous state

The Five Elements of MDP

Tips explained:

State Value Function

Tips:

In short: weighted value!!!

Bellman's equation

The core of the Bellman equation: the value of the current state = the current reward + the value of the next step