Objectives
There have been many recent advances in the field of reinforcement learning. The objective of the course is to provide exposure to these advances and facilitate in depth discussions on chosen topics. The course requires that students have done the introduction to reinforcement learning course.
Course contents
- Introduction: Recent Advances in Reinforcement Learning – Atari Game Player, Alpha Go, and other case studies.
- Model Based RL: Bayesian Approaches to Reinforcement Learning; Data-efficient Reinforcement Learning; Learning with off-line data; Learning with incompletely specified models; RL and planning.
- Human in the Loop RL: Learning with human supervision; imitation learning; inverse reinforcement learning; learning from demonstration.
- Representation Learning for RL: Deep Q network; Deep Actor-Critic; Representation and policy transfer in RL.
- Hierarchical RL: Hierarchical frameworks; Option discovery; safe state abstractions; hierarchies for transfer.
Text Books:
- Richard S. Sutton and Andrew G. Barto. Introduction to Reinforcement Learning, 2nd Edition, MIT Press. 2017. [Draft copies available now]
- Neuro Dynamic Programming. Dimitri Bertsikas and John G. Tsitsiklis. Athena Scientific. 1996.
