Navigating to the Best Policy in Markov Decision Processes.
Aymen Al MarjaniAurélien GarivierAlexandre ProutièrePublished in: NeurIPS (2021)
Keyphrases
- markov decision processes
- optimal policy
- policy iteration
- markov decision process
- average reward
- finite horizon
- infinite horizon
- partially observable
- average cost
- state and action spaces
- action space
- finite state
- state space
- reinforcement learning
- decision processes
- decision problems
- reward function
- policy evaluation
- dynamic programming
- transition matrices
- discount factor
- total reward
- expected reward
- markov decision problems
- partially observable markov decision processes
- policy iteration algorithm
- discounted reward
- stationary policies
- state dependent
- multistage
- decision theoretic planning
- long run
- risk sensitive
- planning under uncertainty
- approximate dynamic programming
- factored mdps
- control policies
- reachability analysis
- model based reinforcement learning
- reinforcement learning algorithms
- sufficient conditions
- action sets
- continuous state
- semi markov decision processes