Policy iteration: for want of recursive feasibility, all is not lost.
Mathieu GranzottoOlivier Lindamulage De SilvaRomain PostoyanDragan NesicZhong-Ping JiangPublished in: CoRR (2022)
Keyphrases
- policy iteration
- markov decision processes
- model free
- fixed point
- least squares
- optimal policy
- reinforcement learning
- finite state
- sample path
- temporal difference
- policy evaluation
- markov decision process
- average reward
- infinite horizon
- linear programming
- optimal control
- markov decision problems
- convergence rate
- discounted reward
- function approximation
- dynamic programming
- data mining
- markov chain
- state space
- pairwise
- optimal solution