Nearly Minimax Optimal Reinforcement Learning for Linear Markov Decision Processes.
Jiafan HeHeyang ZhaoDongruo ZhouQuanquan GuPublished in: ICML (2023)
Keyphrases
- markov decision processes
- reinforcement learning
- dynamic programming
- average cost
- average reward
- optimal policy
- action sets
- finite horizon
- state space
- finite state
- reinforcement learning algorithms
- policy iteration
- state and action spaces
- total reward
- discounted reward
- model based reinforcement learning
- optimal control
- action space
- transition matrices
- stationary policies
- decision theoretic planning
- factored mdps
- policy iteration algorithm
- planning under uncertainty
- optimality criterion
- function approximators
- stochastic games
- infinite horizon
- partially observable
- function approximation
- reachability analysis
- machine learning
- reward function
- temporal difference
- optimal strategy
- continuous state spaces
- expected reward
- markov decision process
- control policies
- action selection
- model free
- decision problems
- linear programming
- optimal solution
- learning algorithm