IMED-RL: Regret optimal learning of ergodic Markov decision processes.
Fabien PesquerelOdalric-Ambrym MaillardPublished in: NeurIPS (2022)
Keyphrases
- markov decision processes
- reinforcement learning
- dynamic programming
- state space
- optimal policy
- policy iteration
- learning algorithm
- partially observable
- total reward
- average cost
- finite state
- model based reinforcement learning
- reinforcement learning algorithms
- action sets
- continuous state spaces
- average reward
- stochastic games
- reward function
- infinite horizon
- finite horizon
- state abstraction
- transition matrices
- action space
- optimal control
- actor critic
- markov decision process
- decision theoretic planning
- discounted reward
- real time dynamic programming
- search space
- rl algorithms
- supervised learning
- reachability analysis