Deep Robust Reinforcement Learning and Regularization

Explore deep robust reinforcement learning and regularization in this 31-minute lecture by Shie Mannor from Technion. Delve into classical reinforcement learning concepts before examining the meaning of robustness and three types of uncertainties. Investigate planning with parameter uncertainty, robust MDPs, and robust policy evaluation. Learn about action robustness, posterior uncertainty sets, and the uncertainty robust Bellman equation. Conclude with insights on deep learning approximation in the context of robust reinforcement learning.

Intro
Classical Reinforcement Learning
Motivation
Meaning of robustness
Three Types of Uncertainties
Applicability
Introduction - Planning with Parameter Uncertainty
Background: Robust MDPS
Robust Policy Evaluation
Part 2
Action Robustness
Some results
Posterior Uncertainty Sets: Online Construction of Uncertainty Sets
Uncertainty Robust Bellman Equation
Deep Learning Approximation
Conclusion