Explore hierarchical reinforcement learning through an in-depth examination of FeUdal Networks in this 21-minute lecture presented by Rene Bidart. Delve into the architecture of FeUdal Networks, understanding the roles of manager and worker components, and how they interact to solve complex tasks. Investigate the use of dilated LSTM and its impact on temporal resolution. Analyze the application of FeUdal Networks in challenging environments like the Water Maze, and discover how intrinsic rewards and transfer learning contribute to improved performance. Gain valuable insights into the potential and limitations of this approach, concluding with a summary and thought-provoking discussion on the future of hierarchical reinforcement learning.
Overview
Syllabus
Intro
Hierarchies
FeUdalism
Architecture
Manager
Worker
Dilated LS TM
Results
Water Maze
Temporal Resolution
Intrinsic Reward
Transfer
Summary
Thoughts
Taught by
Pascal Poupart