Bartsekas D. Class Notes for Topics in Reinforcement Learn. 2022

Textbook in PDF format

The purpose of the notes is to provide an entry point to reinforcement learning, largely from a decision, control, and optimization point of view. They have limited scope, but they provide enough background for starting to read literature in the field and for making a choice for a research-oriented term paper. They roughly cover the material of the first six to seven lectures. They also provide the foundation for much of the material of the remaining lectures, which address topics such as parametric cost function and policy approximations (possibly involving neural networks), approximation in policy space, and aggregation, and deal much more broadly with infinite horizon problems. To repeat, the notes (and associated slides) are not a textbook, they just provide an entry point to the field, often uneven in coverage and insufficient for in-depth understanding. At the same time the notes are at the forefront of current research, and in part discuss new and as yet unpublished material, which is also covered in my recent textbooks