Provably Efficient Reinforcement Learning for Infinite-Horizon Average-Reward Linear MDPs.
Kihyuk HongYufan ZhangAmbuj TewariPublished in: CoRR (2024)
Keyphrases
- average reward
- markov decision processes
- infinite horizon
- optimal policy
- reinforcement learning
- policy iteration
- finite horizon
- long run
- state space
- total reward
- stochastic games
- markov decision process
- average cost
- state and action spaces
- partially observable
- semi markov decision processes
- discount factor
- dynamic programming
- reinforcement learning algorithms
- decision problems
- finite state
- discounted reward
- planning under uncertainty
- markov decision problems
- optimal control
- multistage
- function approximation
- model free
- state action
- actor critic
- reward function
- policy gradient
- decision making
- least squares
- initial state
- learning algorithm