PAC Bounds for Multi-armed Bandit and Markov Decision Processes.
Eyal Even-DarShie MannorYishay MansourPublished in: COLT (2002)
Keyphrases
- markov decision processes
- multi armed bandit
- reinforcement learning
- upper bound
- regret bounds
- discounted reward
- state space
- finite state
- lower bound
- optimal policy
- transition matrices
- dynamic programming
- sample complexity
- upper and lower bounds
- policy iteration
- sample size
- average reward
- function approximation
- average cost
- decision theoretic planning
- state and action spaces
- markov decision process
- partially observable
- action space
- learning algorithm
- infinite horizon
- online learning
- worst case
- model free
- optimal control
- learning tasks
- least squares
- special case
- multi agent
- machine learning