Towards Tight Bounds on the Sample Complexity of Average-reward MDPs.
Yujia JinAaron SidfordPublished in: CoRR (2021)
Keyphrases
- tight bounds
- average reward
- markov decision processes
- upper bound
- optimal policy
- semi markov decision processes
- long run
- reinforcement learning
- policy iteration
- stochastic games
- discounted reward
- optimality criterion
- finite state
- state space
- state and action spaces
- model free
- markov chain
- decision theoretic planning
- state action
- factored mdps
- dynamic programming
- infinite horizon
- planning under uncertainty
- policy gradient
- lower bound
- partially observable
- decision problems
- dynamical systems
- sufficient conditions
- linear programming
- discount factor
- computational complexity