Asymptotic linear programming and policy improvement for singularly perturbed Markov decision processes.
Eitan AltmanKonstantin AvrachenkovJerzy A. FilarPublished in: Math. Methods Oper. Res. (1999)
Keyphrases
- markov decision processes
- linear programming
- optimal policy
- policy iteration
- dynamic programming
- markov decision problems
- markov decision process
- average cost
- state and action spaces
- infinite horizon
- finite horizon
- asymptotically optimal
- state space
- average reward
- transition matrices
- partially observable
- action space
- reward function
- decision processes
- decision theoretic planning
- finite state
- linear program
- reinforcement learning
- markov games
- discounted reward
- policy evaluation
- reachability analysis
- multistage
- policy iteration algorithm
- factored mdps
- stationary policies
- optimal solution
- model based reinforcement learning
- planning under uncertainty
- decision problems
- reinforcement learning algorithms
- partially observable markov decision processes
- expected reward
- total reward
- objective function
- state dependent
- long run
- markov chain
- actor critic
- action sets
- real valued
- semi markov decision processes
- action selection
- machine learning