Explore trust region methods and proximal policy optimization in this 22-minute video lecture from the CS885 course at the University of Waterloo. Delve into gradient policy optimization, Kullback-Leibler Divergence, and the Trust Region Policy Optimization (TRPO) algorithm. Learn about constrained optimization and the simplified objective of Proximal Policy Optimization (PPO). Examine empirical results and illustrations to reinforce your understanding of these advanced reinforcement learning concepts. Access accompanying slides from the course website for a comprehensive learning experience.
Overview
Syllabus
Gradient policy optimization
Recall Policy Gradient
Trust region method
Trust region for policies
Kullback-Leibler Divergence
Reformulation
Derivation (continued)
Trust Region Policy Optimization (TRPO) TRPOO Initialize sa to anything Loop forever (for each episode)
Constrained Optimization
Simpler Objective
Proximal Policy Optimization (PPO)
Empirical Results
Illustration
Taught by
Pascal Poupart
Related Courses
-
Deep Reinforcement Learning in Python
-
Trust Region Policy Optimization
-
Reinforcement Learning: From Policy Optimization to Multi-Agent Systems
-
Trust Region Based Adversarial Attack on Neural Networks
-
Overview of Deep Reinforcement Learning Methods
-
Deep Reinforcement Learning of Marked Temporal Point Processes
