Policy Regularization with Dataset Constraint for Offline Reinforcement Learning.
Yuhang RanYi-Chen LiFuxiang ZhangZongzhang ZhangYang YuPublished in: ICML (2023)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- action selection
- partially observable environments
- control policy
- reinforcement learning problems
- markov decision problems
- markov decision process
- function approximation
- approximate dynamic programming
- temporal difference
- state space
- policy iteration
- reinforcement learning algorithms
- markov decision processes
- linear constraints
- reward function
- function approximators
- control policies
- policy gradient
- policy evaluation
- state action
- actor critic
- dynamic programming
- learning process
- state and action spaces
- partially observable domains
- partially observable
- long run
- regularization parameter
- benchmark datasets
- learning algorithm
- machine learning
- rl algorithms
- regularization term
- decision problems
- transfer learning
- image restoration
- feature set
- supervised learning
- constrained minimization
- conjugate gradient algorithm
- real time