Markov Decision Processes 2 - Reinforcement Learning | Stanford CS221: AI (Autumn 2019)

Рет қаралды 69,079

Күн бұрын

For more information about Stanford’s Artificial Intelligence professional and graduate programs, visit: stanford.io/2Zv1JpK
Topics: Reinforcement learning, Monte Carlo, SARSA, Q-learning, Exploration/exploitation, function approximation
Percy Liang, Associate Professor & Dorsa Sadigh, Assistant Professor - Stanford University
onlinehub.stanford.edu/
Associate Professor Percy Liang
Associate Professor of Computer Science and Statistics (courtesy)
profiles.stanford.edu/percy-l...
Assistant Professor Dorsa Sadigh
Assistant Professor in the Computer Science Department & Electrical Engineering Department
profiles.stanford.edu/dorsa-s...
To follow along with the course schedule and syllabus, visit:
stanford-cs221.github.io/autu...

Пікірлер: 9

@henkjekel4081 Жыл бұрын

Yeah, u really need to be having an episode to play this game

@aojing 3 ай бұрын

A legacy question from last MDP-1 is still hovering around 2: What is the Transition function for this class? Is it a function of Action?

@inventwithdean Ай бұрын

It is a function of both State and Action.

@black-sci 4 ай бұрын

Somehow Lecture left me confused in the end. may be I should rewatch.

@JumbyG Жыл бұрын

I think there may be a typo at 28:27, it states that the Qpi is (4+8+16)/3 however I believe it should be (4+8+12)/3? Please correct me if I am wrong

@seaotterlabs1685 Жыл бұрын

I think it should be (4+8+16)/3, as I believe their last run has four 4 values.

@endoumamoru3835 6 ай бұрын

he is calculating sum of all rewards you can get. First time sum was 4 as only one reward was present and next was 8 as 2 rewards and then next it was 16 as 4 rewards were there.

@albert2266 2 ай бұрын

Just to clarify a concept as I think 7:29 is not true because value function shouldn't be equal to the Q value. Value function is the expected utility for "all possible actions" at a given state. Therefore, it should be the expected Q_pi rather than just simply equal to Q_pi since Q_pi is the expected utility for "a given action" at a given state. Please correct me if I'm wrong.