Offline Reinforcement Learning with Instrumental Variables in Confounded Markov Decision Processes.
Zuyue FuZhengling QiZhaoran WangZhuoran YangYanxun XuMichael R. KosorokPublished in: CoRR (2022)
Keyphrases
- markov decision processes
- reinforcement learning
- state space
- optimal policy
- reinforcement learning algorithms
- policy iteration
- finite state
- action space
- infinite horizon
- decision theoretic planning
- model based reinforcement learning
- partially observable
- state and action spaces
- reachability analysis
- state variables
- finite horizon
- dynamic programming
- planning under uncertainty
- markov decision process
- transition matrices
- function approximation
- action sets
- factored mdps
- stochastic games
- state abstraction
- average cost
- multi agent
- markov chain
- monte carlo
- policy evaluation
- average reward
- continuous state
- model free
- action selection
- temporal difference
- total reward
- reward function
- control problems
- machine learning