Pessimism in the Face of Confounders: Provably Efficient Offline Reinforcement Learning in Partially Observable Markov Decision Processes.
Miao LuYifei MinZhaoran WangZhuoran YangPublished in: CoRR (2022)
Keyphrases
- reinforcement learning
- partially observable markov decision processes
- continuous state
- partially observable domains
- partial observability
- optimal policy
- belief state
- state space
- markov decision processes
- finite state
- hidden state
- function approximation
- learning algorithm
- policy gradient
- partially observable environments
- dynamical systems
- domain specific
- model free
- belief space
- partially observable
- decision problems
- knowledge base