Course Overview
Graduate-level course covering modern RL techniques including offline RL, safe RL, and applications to real-world problems. Emphasis on bridging theory and practice with hands-on projects.
Course Description
This course provides an in-depth exploration of advanced reinforcement learning topics, with a focus on techniques and challenges relevant to real-world deployment. Students will gain both theoretical understanding and practical experience through implementation projects.
Learning Objectives
By the end of this course, students will be able to:
- Understand fundamental RL algorithms and their theoretical properties
- Implement and debug modern RL algorithms from scratch
- Identify appropriate RL formulations for real-world problems
- Recognize and address practical challenges in RL deployment
- Critically evaluate RL research papers and reproduce key results
Tentative Course Outline
Part I: Foundations (Weeks 1-4)
- Markov Decision Processes and Dynamic Programming
- Policy Gradient Methods
- Value-Based Methods (Q-Learning, DQN)
- Actor-Critic Algorithms
Part II: Advanced Methods (Weeks 5-8)
Offline Reinforcement Learning
- Conservative Q-Learning
- Behavior regularization
- Dead-end identification
Safe Reinforcement Learning
- Constrained MDPs
- Risk-sensitive RL
- Uncertainty quantification
Transfer Learning and Meta-Learning
- Domain adaptation
- Multi-task RL
- Few-shot RL
Part III: Applications (Weeks 9-12)
- RL for Language Models
- Healthcare applications
- Robotics and control
- Real-world deployment considerations
Part IV: Research Projects (Weeks 13-16)
- Student-led research projects
- Paper presentations
- Final project presentations
Prerequisites
Required:
- Strong programming skills (Python)
- Probability and statistics
- Linear algebra
- Basic machine learning (equivalent to an introductory ML course)
Recommended:
- Prior exposure to deep learning
Course Format
- Lectures: 2x weekly, covering theory and algorithms
- Lab Sessions: Weekly hands-on implementation exercises
- Paper Discussions: Student-led presentations of recent papers
- Projects: Individual or team projects applying RL to real problems
Assessment
- Homework assignments (30%): Implementation-focused problem sets
- Paper presentations (20%): Present and lead discussion on research papers
- Midterm exam (20%): Covering fundamental concepts
- Final project (30%): Original research or substantial application
Recommended Resources
Textbooks
- Sutton & Barto: “Reinforcement Learning: An Introduction” (2nd ed)
- Bertsekas: “Dynamic Programming and Optimal Control”
Papers & Surveys
- Curated reading list of foundational and recent papers
- Survey papers on offline RL, safe RL, and applications
Practical Resources
- OpenAI Gym / Gymnasium
- Stable-Baselines3
- D4RL benchmark suite
Coming Soon
Detailed syllabus, lecture notes, and course materials will be made available as the course develops. Stay tuned!