CVPR 2018 Tutorial on Inverse Reinforcement Learning for Computer Vision

Friday, June 22nd. Ball C (400). 8:45am - 12:00pm. Exact timing TBA.


Modern computer vision is great at analyzing pictures, but what about the people and robots in the pictures? What are their motivations? What will they do in the future? Are they behaving optimally? Inverse Reinforcement Learning (IRL) is well-suited to answer these kinds of questions. IRL provides a framework to learn and reason about the intent underlying goal-driven behavior. While there have been some successes in applying IRL to computer vision problems, we believe that IRL as a tool is underappreciated by computer vision researchers seeking to understand goal-driven behavior. In this tutorial, we will give an overview of the motivation, application, and practical aspects of IRL as applied to computer vision problems. The tutorial will be as self-contained as possible and will cover the relevant background material on Reinforcement Learning (RL).


Nick Rhinehart (CMU)
Kris Kitani (CMU)
Paul Vernaza (NECLA)


08:45 - 09:00 Introduction: What is IRL? What is RL?
09:00 - 09:15 Successes of IRL
09:15 - 10:00 A Reinforcement Learning Primer
10:00 - 10:30 Coffee Break
10:30 - 11:00 Inverse Reinforcement Learning I
11:00 - 11:30 Inverse Reinforcement Learning II
11:30 - 12:00 Applications to Computer Vision In Real Life (IRL IRL)


We will provide presentation materials here, check back later.