Near Optimality of Finite Memory Feedback Policies in Partially Observed Markov Decision Processes.
Ali Devran KaraSerdar YükselPublished in: J. Mach. Learn. Res. (2022)
Keyphrases
- markov decision processes
- partially observed
- stationary policies
- average cost
- optimal policy
- average reward
- state and action spaces
- finite state
- action sets
- state space
- markov decision process
- finite number
- reinforcement learning
- policy iteration
- decision processes
- total reward
- infinite horizon
- expected reward
- dynamic programming
- reinforcement learning algorithms
- linear program
- decision theoretic planning
- decentralized control
- finite horizon
- reward function
- long run
- discounted reward
- model based reinforcement learning
- risk sensitive
- partially observable markov decision processes
- transition matrices
- policy iteration algorithm
- factored mdps
- control policy
- planning under uncertainty
- lot sizing
- sufficient conditions
- control policies
- macro actions
- decision problems
- reachability analysis
- probabilistic planning
- initial state
- semi markov decision processes
- np hard
- optimal solution