Course Topics
The course focuses on general deep reinforcement learning (deep RL) techniques, algorithms, and applications, including off-policy RL, on-policy RL, model-based RL, imitation learning, sim-to-real, curriculum learning, safety-aware RL, and scalable RL. This course requires good knowledge in machine learning, deep learning, and statistical methods. The course can be further synergetic with knowledge in robotics, control, and physics-based simulation.
Instructor
Instructor: Sehoon Ha (sehoonha@gatech.edu, Klaus 3326)
Co-instructor: Jie Tan (jietan@google.com, remote teaching)
Teaching Assistant: Ren Liu (rliu384@gatech.edu)
Class Sessions
3 Credit Hours
Hybrid-mode (In-person + streaming)
11:00 am - noon MWF at Molecular Sciences and Engr 1224 Scheller College of Business 223
No Textbook
Grading
Term project: 60% (= Proposal 20% + Midterm 20% + Final 20%)
Paper presentation: 30%
Class participation: 10%
Lecture
The instructors will give introductory lectures for each topic, which are typically followed by a couple of paper presentation sessions. Each lecture will typically span a single class and give a brief overview of mathematical and algorithmic backgrounds.
Special Lecture
There will also be two to three special lectures. The potential topics include guest speakers, paper writing, or debates on tricky subjects.
Term Project
The main portion of the class is the term project. Two to three students will formulate a team and conduct a research project over the semester. The expectation is to implement a reasonable baseline algorithm and improve it based on the team’s own idea. The projects will be guided by the instructors via proposal, midterm, and final presentation sessions. Each team will also submit a midterm report (2 pages) and a final report (4 pages). The successful projects will be discussed to submit to international conferences.
Paper Presentation
Each student will present a paper. The presentation is expected to be 12 minutes, which will be followed by 10 minutes of discussions. The students will sign up for the paper presentation at the beginning of the semester.
Class Participation
Class presence and participation points (10%) are given to encourage your active class participation and discussion. If necessary, please send the instructor a brief email to explain their absence in advance.
Final Note
The class is designed for highly motivated students who want to start deep reinforcement learning and robotics research. The instructors’ expectations can be high.
COVID-19 Related Statement
The instructor will try best to follow the school's policy (https://health.gatech.edu/tech-moving-forward), which include:
will try to conduct in-person teaching, but will not mandate the student's physical attendance.
will not mandate/ask for vaccination or masks (personally encourage both).
will try to provide the best remote learning experience as much as possible to accommodate various situations.
please contact the instructor (sehoonha@gatech.edu) for any questions.
