PESSIMISTIC BOOTSTRAPPING FOR UNCERTAINTY-DRIVEN OFFLINE REINFORCEMENT LEARNING

Keyphrases

• Bootstrap 100%
• Offline Reinforcement Learning 100%
• Offline RL 42%
• Value Function 28%
• Distribution Sampling 28%
• Learning Objectives 14%
• Uncertainty Quantification 14%
• Sampling Methods 14%
• Function-based 14%
• Offline Algorithm 14%
• Uncertainty Estimation 14%
• Policy Constraints 14%
• Behavioral Policy 14%
• Q-function 14%
• Offline Data 14%
• Out-of-distribution Data 14%
• Extrapolation Error 14%

Computer Science

• Reinforcement Learning 100%
• Function Value 50%
• Sampling Distribution 50%
• Data Distribution 25%