This course aims at studying basic theory and practical algorithms of reinforcement learning (RL).
By the end of the course, students are expected to
define and explain key features of RL,
know how to use RL for a given application problem,
implement common RL algorithms,
understand theoretical and empirical approaches for evaluating the quality of a RL algorithm, and
hopefully, formulate and solve research problems in RL.

Mandatory: AI, machine learning, calculus, probability & statistics
Recommend: optimization, programming

The grade is based on
quizzes and class participation (30%),
assignments (30%),
paper presentation (10%), and
project (30%).
If you miss five classes, you will receive F, no matter what.

There is no official textbook.
Some references:
- “Reinforcement Learning: An Introduction”, by R. S. Sutton and A. G. Barto, MIT Press, 2020. (http://incompleteideas.net/book/RLbook2020.pdf)
- “Bandit algorithm” by T. Lattimore and C. Szepesvari

1. Introduction
2. MDP
3. Model-Free Evaluation & Control
4. Policy Gradient, PPO
5. Imitation Learning
6. RLHF and DPO
7. Offline RL
8. Multi-Armed Bandit
9. Bayesian Bandit
10. Data Efficient RL
11. Monte-Carlo Tree Search
12. Case studies: AlphaGo, DDPG, GRPO, etc
13. Research paper presentations

Offline.

.