A Linearly Relaxed Approximate Linear Program for Markov Decision Processes.
Chandrashekar LakshminarayananShalabh BhatnagarCsaba SzepesváriPublished in: CoRR (2017)
Keyphrases
- linear program
- markov decision processes
- factored mdps
- dynamic programming
- optimal solution
- linear programming
- average cost
- approximate dynamic programming
- policy evaluation
- optimal policy
- policy iteration
- stationary policies
- finite state
- state space
- reinforcement learning
- semi infinite
- column generation
- transition matrices
- partially observable
- stochastic programming
- model based reinforcement learning
- average reward
- planning under uncertainty
- multistage
- decision theoretic planning
- approximate solutions
- finite horizon
- np hard
- action space
- markov decision process
- objective function
- exact solution
- machine learning
- search algorithm
- context specific
- decision processes
- infinite horizon