Discovery of Optimal Solution Horizons in Non-Stationary Markov Decision Processes with Unbounded Rewards.
Grigory NeustroevMathijs de WeerdtRemco A. VerzijlberghPublished in: ICAPS (2019)
Keyphrases
- markov decision processes
- non stationary
- optimal solution
- infinite horizon
- finite state
- state space
- optimal policy
- reinforcement learning
- transition matrices
- dynamic programming
- finite horizon
- decision theoretic planning
- search space
- policy iteration
- planning under uncertainty
- action space
- reachability analysis
- sequential decision making under uncertainty
- reinforcement learning algorithms
- average reward
- empirical mode decomposition
- expected reward
- np hard
- data mining
- linear program
- partially observable
- factored mdps
- objective function
- random fields
- model based reinforcement learning
- state abstraction
- real time dynamic programming
- action sets
- linear programming
- reward function
- markov decision process
- average cost