Pessimistic Q-Learning for Offline Reinforcement Learning: Towards Optimal Sample Complexity.
Laixi ShiGen LiYuting WeiYuxin ChenYuejie ChiPublished in: ICML (2022)
Keyphrases
- reinforcement learning
- sample complexity
- sequential decision problems
- learning algorithm
- learning problems
- function approximation
- supervised learning
- dynamic programming
- state space
- reinforcement learning algorithms
- optimal control
- markov decision processes
- pac learning
- multi agent
- optimal policy
- theoretical analysis
- model free
- generalization error
- vc dimension
- upper bound
- lower bound
- action selection
- control policy
- multi agent reinforcement learning
- active exploration
- continuous state and action spaces
- machine learning
- temporal difference
- active learning
- sample size
- worst case
- special case
- continuous state spaces
- training set
- objective function
- single agent
- function approximators
- small number
- markov decision problems
- optimal solution