強化學習專論
Selected Topics in Reinforcement Learning
| 節 | 週二 |
|---|---|
A 18:30–19:20 | 強化學習專論 EC114 3 節連堂 |
B 19:30–20:20 | |
C 20:30–21:20 |
* 根據陽明交大上課時間表所列
This course is designed to teach students how to develop reinforcement learning (RL) algorithms for a wide range of applications, such as computer games, video games, intelligent traffic management, manufacturing scheduling, autonomous driving/racing, and robotics. The objectives are summarized as follows. (1) To understand the basic core concepts of reinforcement learning (RL) (2) To understand many latest RL techniques for applications (3) To familiarize with tools for developing RL, such as PyTorch, Gazebo, etc. (4) To develop practical working systems via projects such as DeepRacer.
Machine Learning/Deep Learning (suggested)
Projects (done individually) 50% Paper presentation (done in groups of 2 members) 20% Final exam 30%
- Core of RL
- Advanced Topics of RL
- Presentation
- Introduction to RL
| 週次 | 主題 |
|---|---|
| 第 1 週 | Introduction to Reinforcement Learning |
| 第 2 週 | Case studies of lightweight model applications: 2048 and Go |
| 第 3 週 | Fundamentals: Markov Decision Process (MDP), Dynamic Programming (Tabular RL), Q-Learning, Function Approximation |
| 第 4 週 | Value-Based Reinforcement Learning: DQN, DDQN (Double DQN), Dueling Network (with Advantage), Distributional DQN |
| 第 5 週 | Policy-based Reinforcement Learning: Policy Gradient, Actor-Critic (Discrete actions), A2C and A3C (Asynchronous Advantage Actor-Critic) |
| 第 6 週 | Policy-based Reinforcement Learning: TRPO & PPO, GAE, DDPG, TD3 (Continuous Actions), SAC (Soft Actor-Critic) |
| 第 7 週 | Applications: DeepRacer: Augmentation, RL-cycleGAN, DrQ |
| 第 8 週 | Applications: Solving Rubik Cube, RL for optimization (JSP/TSP) |
| 第 9 週 | Exploration vs. Exploitation: Multi-Arm Bandits, UCB, Sequential Halving |
| 第 10 週 | Planning: Dyna, Monte-Carlo Tree Search (MCTS), AlphaGo, AlphaZero, MuZero, Path Consistency, Abstraction |
| 第 11 週 | Advanced Exploration: ICM, RND Experience Replay: PER, Ape-X |
| 第 12 週 | Model-based RL: DQfD, R2D3 Multi-Agents RL (MARL) Q-mix, COMA |
| 第 13 週 | Presentation |
| 第 14 週 | Presentation |
| 第 15 週 | Presentation |
| 第 16 週 | Final exam |
| 第 17 週 | Final competition for DeepRacer |
| 第 18 週 |
1. R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, Nov. 2017 2. David Silver, Online Course for Deep Reinforcement Learning. http://www.cs.ucl.ac.uk/staff/D.Silver/web/Teaching.html 3. Papers and slides.
- 地點
- TBA
- 時間
- TBA
- 聯絡方式
- TBA