[CS筆記01] CS234 Winter 2019 1-2

1. Markov Reward Process

1) Closed-Form

2) DP

2. Markov Decision Process
1) Policy Evaluation

2) Policy Iteration

- CS234 (Compute infinite horizon value of a policy!)

- MIT 16.410

3) State-Action Value Q 

Q function basically tells you how good it is to be in state S and perform action A, and follow policy pi from the next state onwards.

4) Policy Improvement

5) Value Iteration

Compute optimal value for horizon = k!

6) Bellman backup operator