An unbounded Berge's minimum theorem with applications to discounted Markov decision processes.
Raúl Montes-de-OcaEnrique Lemus-RodríguezPublished in: Kybernetika (2012)
Keyphrases
- markov decision processes
- optimal policy
- finite state
- state space
- reinforcement learning
- infinite horizon
- dynamic programming
- transition matrices
- finite horizon
- average cost
- policy iteration
- average reward
- decision theoretic planning
- action space
- reinforcement learning algorithms
- decision processes
- action sets
- partially observable
- reachability analysis
- state abstraction
- risk sensitive
- markov decision process
- decision diagrams
- factored mdps
- model based reinforcement learning
- discounted reward
- decision problems
- learning algorithm
- discount factor
- expected reward
- multistage