Offline Reinforcement Learning with On-Policy Q-Function Regularization.
Laixi ShiRobert DadashiYuejie ChiPablo Samuel CastroMatthieu GeistPublished in: CoRR (2023)
Keyphrases
- reinforcement learning
- function approximators
- optimal policy
- policy gradient
- control policy
- function approximation
- state action
- policy search
- action selection
- policy iteration
- temporal difference
- real time
- neural network
- partially observable
- approximate dynamic programming
- learning process
- learning algorithm
- learning problems
- reproducing kernel hilbert space
- markov decision processes
- markov decision process
- action space
- machine learning
- rl algorithms
- transition model
- reinforcement learning problems
- smoothing parameter
- continuous state
- control policies
- reward function
- model free
- regularization parameter
- learning tasks
- decision problems
- state space