The course will provide an in-depth view on the state-of-the-art learning methods for control and the know-how to apply these techniques. The first half of the course will focus on hand-on experience through exercises. The second half will focus on current research directions and open questions. Topics will span reinforcement learning, self-supervised learning, imitation learning, model-based learning and advanced deep learning architectures. By the end of the course, we hope you will be able to answer if learning based control can help solve the problem of your interest, how to formulate the problem in the learning framework and what algorithms to use. It will also prepare you for research in this area.
This is a graduate course that will be a mix of seminar and lecture style classes. The plan is delve into technical details of particular algorithmic topics and follow it up with reading research papers. The course will involve working on a research project, three assignments and presenting papers.
- Working knowledge of machine learning, deep learning, reinforcement learning is assumed.
- Experience with deep learning packages such as PyTorch or TensorFlow is assumed. Homeworks will be in Python.
- This is NOT a basic course in reinforcement learning, deep learning or AI. Although, this can be used as a concentration subject in AI.
Time: Tuesday/Thursday 11am-12:30pm
Location: virtual (zoom)
|Pulkit Agrawal||Tuesday 12:30pm-1:10pm|
|Tao Chen||Monday 5pm-6pm|
|Joshua Gruenstein||Wednesday 4pm-5pm|
Forums are on Piazza (sign up with your MIT email address). Access Code for piazza will be given in the first lecture. You can also email the instructors for the access code. Please checkout the piazza regularly, we will make new annoncements on piazza.
Assignments are due one week after the assignment release. Late assignment submissions will be penalized 10% every 24 hours. You will also need to submit a brief status report every week (on Gradescope), contributing to the class participation grade.