Discounted continuous-time Markov decision processes with unbounded rates and randomized history-dependent policies: the dynamic programming approach.
Alexey B. PiunovskiyYi ZhangPublished in: 4OR (2014)
Keyphrases
- markov decision processes
- dynamic programming
- stationary policies
- optimal policy
- state space
- optimal control
- markov decision process
- infinite horizon
- finite state
- policy iteration algorithm
- action sets
- average cost
- total reward
- average reward
- discounted reward
- decision processes
- reward function
- policy iteration
- finite horizon
- markov decision problems
- reinforcement learning
- control policies
- planning under uncertainty
- transition matrices
- reachability analysis
- decision problems
- action space
- long run
- multistage
- markov chain
- macro actions
- model based reinforcement learning
- decentralized control
- linear program
- reinforcement learning algorithms
- factored mdps
- decision theoretic planning
- linear programming
- discount factor
- partially observable markov decision processes
- expected reward
- machine learning
- state abstraction
- inventory level
- semi markov decision processes
- partially observable
- learning algorithm