Dynamic Regret of Online Markov Decision Processes.
Peng ZhaoLongfei LiZhi-Hua ZhouPublished in: CoRR (2022)
Keyphrases
- markov decision processes
- online learning
- finite state
- state space
- reward function
- total reward
- reinforcement learning
- optimal policy
- policy iteration
- transition matrices
- model based reinforcement learning
- dynamic programming
- decision theoretic planning
- planning under uncertainty
- finite horizon
- infinite horizon
- reinforcement learning algorithms
- decision processes
- risk sensitive
- reachability analysis
- average reward
- average cost
- partially observable
- factored mdps
- action space
- dynamic environments
- state and action spaces
- expected reward
- action sets
- online convex optimization
- least squares
- lower bound