Instance-Dependent Near-Optimal Policy Identification in Linear MDPs via Online Experiment Design.

Andrew Wagenmaker Kevin G. Jamieson

Published in: NeurIPS (2022)

Keyphrases

optimal policy
markov decision processes
finite horizon
state space
reinforcement learning
decision problems

dynamic programming
average cost
markov decision problems
markov decision process
long run
multistage

infinite horizon
policy iteration
average reward
dynamic programming algorithms
initial state

lost sales
develop a mathematical model
sufficient conditions
state dependent
linear programming