Explore offline reinforcement learning in this 40-minute discussion moderated by Pablo Castro from Google. Delve into topics such as small vs large data sets, optimism under uncertainty, state coverage, and pragmatic vs conceptual approaches. Examine online R, policy evaluation, assumptions, and various types of uncertainty qualification and quantification. Investigate methods without constraints, offline data, guidelines for collecting data, and gain insights from expert answers to thought-provoking questions in the field of deep reinforcement learning.
Discussion - Offline Reinforcement Learning
Overview
Syllabus
Intro
Small data sets vs large data sets
Optimism under uncertainty
State coverage
Pragmatic approach
Conceptual approach
Online R
Policy Evaluation
I dont know
Assumptions
Uncertainty qualification
Uncertainty types
Uncertainty quantification
Scotts question
Scotts answer
Javas answer
Emmas answer
Higher level question
Methods without constraints
Offline data
Guidelines
Collecting data
Conclusion
Taught by
Simons Institute
Related Courses
-
Generalization and Robustness in Offline Reinforcement Learning
-
What Are the Statistical Limits of Offline Reinforcement Learning With Function Approximation?
-
Batch Offline Reinforcement Learning - Part 1
-
Offline Reinforcement Learning and Model-Based Optimization
-
Reinforcement Learning from Human Feedback - From Zero to ChatGPT
