Deep Q Networks (DQN)
1Notebook can be found at: https://github.com/tanchongmin/TensorFlow-Implementations/tree/main/RL
Unfortunately, we had a lost video for Part 1 on Monte Carlo Tree Search, so that one the only reference is the slides in the github (for now until I re-record another time).
0:00 Introduction
2:50 Recap: Reinforcement Learning
9:00 Recap: Markov Decision Process
13:43 Tabular Q-Learning to Deep Q-Learning
21:33 Neural Network Process for DQN
23:40 Intuition of Q-learning
33:46 Q-Learning as a form of Temporal Difference Learning
38:02 How to generate experience: Epsilon-greedy actions
41:42 How to learn from experience: Replay buffer
46:18 Q-Value
48:27 Learning Q-Values
49:54 Policy & Target Network
54:45 Visualizing Q-Values
56:15 Choosing the Optimal Action
56:44 Improving DQN
1:00:16 Jupyter Notebook on Cart Pole