On the Global Convergence of Policy Gradient in Average Reward Markov Decision Processes.
Navdeep KumarYashaswini MurthyItai ShufaroKfir Y. LevyR. SrikantShie MannorPublished in: CoRR (2024)
Keyphrases
- average reward
- policy gradient
- markov decision processes
- convergence rate
- convergence speed
- policy iteration
- actor critic
- optimal policy
- reinforcement learning
- stochastic games
- state space
- finite state
- state action
- discounted reward
- optimization methods
- dynamic programming
- infinite horizon
- state and action spaces
- reinforcement learning algorithms
- action space
- long run
- partially observable
- gradient method
- markov decision process
- average cost
- learning rate
- step size
- markov chain
- cost function
- evolutionary algorithm
- search algorithm