An envelope theorem and some applications to discounted Markov decision processes.
Hugo Cruz-SuárezRaúl Montes-de-OcaPublished in: Math. Methods Oper. Res. (2008)
Keyphrases
- markov decision processes
- optimal policy
- average reward
- reinforcement learning
- state space
- finite state
- dynamic programming
- planning under uncertainty
- policy iteration
- average cost
- transition matrices
- reachability analysis
- finite horizon
- infinite horizon
- decision theoretic planning
- reinforcement learning algorithms
- partially observable
- factored mdps
- risk sensitive
- decision processes
- markov decision process
- model based reinforcement learning
- action space
- action sets
- semi markov decision processes
- machine learning
- reward function
- decision diagrams
- state and action spaces
- total reward
- discount factor