Nearly Minimax Optimal Regret for Learning Infinite-horizon Average-reward MDPs with Linear Function Approximation.
Yue WuDongruo ZhouQuanquan GuPublished in: CoRR (2021)
Keyphrases
- reinforcement learning
- average reward
- total reward
- function approximation
- markov decision processes
- optimal policy
- infinite horizon
- actor critic
- policy iteration
- stochastic games
- td learning
- finite horizon
- reinforcement learning algorithms
- optimal control
- function approximators
- temporal difference
- partially observable
- policy gradient
- dynamic programming
- long run
- state action
- model free
- average cost
- discounted reward
- learning tasks
- learning algorithm
- action selection
- reward function
- state space
- discount factor
- finite state
- reinforcement learning methods
- supervised learning
- machine learning
- partially observable markov decision processes
- action space
- evaluation function
- rl algorithms
- multistage
- optimal solution
- neural network
- radial basis function
- linear programming
- markov decision problems
- artificial neural networks
- data mining