CS234: Reinforcement Learning Winter 2021

To realize the dreams and impact of AI requires autonomous systems that learn to make good decisions. Reinforcement learning is one powerful paradigm for doing so, and it is relevant to an enormous range of tasks, including robotics, game playing, consumer modeling and healthcare. This class will provide a solid introduction to the field of reinforcement learning and students will learn about the core challenges and approaches, including generalization and exploration. Through a combination of lectures, and written and coding assignments, students will become well versed in key ideas and techniques for RL. Assignments will include the basics of reinforcement learning as well as deep reinforcement learning — an extremely promising new area that combines deep learning techniques with reinforcement learning.

Communication: We will use Piazza for all communications. We encourage all students to use Piazza and you may submit public or private posts.

Learning remotely: We know with covid-19 and other world events, that this quarter will be difficult. To better support your learning during this unprecendented time, we have changed the course format. The new format is experimental, based on the experience of other instructors who have taught recent classes since the pandemic began: please let us know if you have ideas or feedback about how to further improve the class and your learning and we will take those into consideration for this and future offerings.

• Modules (videos and slides): All standard lecture materials will be delivered through modules with pre-recorded course videos that you can watch at your own time. Each week's modules are listed in the schedule and can be accessed here, and will be posted by the end of Sunday before that week's class. Guest lectures will be presented live and recorded for later watching. Recordings will be available to enrolled students through Canvas.
• Lecture Watch Parties: During our class time each Monday, we will have an optional CA-facilitated watch party in breakout rooms for the first lecture in that week. A CA will pause the video at periodic intervals to check your understanding and answer questions. Links to the relevant Zoom or Nook session for a particular watch party will be provided in the schedule.
• Problem sessions: During our class time each Wednesday, we will have an optional CA-lead session to go through exercises about the material and also practice working on materials in breakout rooms with CA support.
• Group office hours: We will have group office hours on Nooks for assignments in addition to 1:1 office hours. Group Office Hours are on 5PM-8PM Wednesday, Thursday, and Friday each week.
• 1:1 office hours: Students can sign up for 1:1 office hours with faculty and CAs. These will all be appointment-based so that students need not to wait in queue. See our calendar for times and sign up links. You can also find the sign up link here. Please contact us through Piazza if you want to schedule one but cannot find an available slot. Video conference links will be provided during sign up.
• Quizzes: Instead of a large high-stakes midterm, there will be four quizzes over the course. We will drop the lowest score of Quiz 1-3.

Platforms: Besides Piazza, we will use a combination of Zoom and Nooks to hold class activities and office hours. All assignments and quizzes will be handled through Gradescope, where you will also find your grades. We will send out links and access codes to enrolled students through Canvas.

Time / Location: All class activities and office hours are in our calendar. Note: Office hour links will be posted by the end of Monday January 11. All times are in Stanford local time (Pacific Time, PT).

You can find previous years (Winter 2020, Winter 2019, Winter 2018) materials here.

By the end of the class students should be able to:

• Define the key features of reinforcement learning that distinguishes it from AI and non-interactive machine learning (as assessed by the exam).
• Given an application problem (e.g. from computer vision, robotics, etc), decide if it should be formulated as a RL problem; if yes be able to define it formally (in terms of the state space, action space, dynamics and reward model), state what algorithm (from class) is best suited for addressing it and justify your answer (as assessed by the exam).
• Implement in code common RL algorithms (as assessed by the assignments).
• Describe (list and define) multiple criteria for analyzing RL algorithms and evaluate algorithms on these metrics: e.g. regret, sample complexity, computational complexity, empirical performance, convergence, etc (as assessed by assignments and the exam).
• Describe the exploration vs exploitation challenge and compare and contrast at least two approaches for addressing this challenge (in terms of performance, scalability, complexity of implementation, and theoretical guarantees) (as assessed by an assignment and the exam).

We believe students often learn an enormous amount from each other as well as from us, the course staff. Therefore to facilitate discussion and peer learning, we request that you please use Piazza for all questions related to lectures and assignments.

For SCPD students, if you have generic SCPD specific questions, please email scpdsupport@stanford.edu or call 650-741-1542. In case you have specific questions related to being a SCPD student for this particular class, please contact us at cs234-win2021-staff@lists.stanford.edu.

For exceptional circumstances that require us to make special arrangements, please email us at cs234-win2021-staff@lists.stanford.edu. For example, such a situation may arise if a student requires extra days to submit a homework due to a medical emergency, or if a student needs to schedule an alternative midterm date due to events such as conference travel etc. They will be considered and approved on a case by case basis.