Towards Minimax Optimal Reward-free Reinforcement Learning in Linear MDPs.
Pihe HuYu ChenLongbo HuangPublished in: ICLR (2023)
Keyphrases
- reinforcement learning
- markov decision processes
- average reward
- dynamic programming
- action sets
- optimal control
- reward function
- total reward
- state space
- optimal policy
- control policy
- model free
- approximate dynamic programming
- function approximation
- function approximators
- policy iteration
- discounted reward
- reinforcement learning algorithms
- eligibility traces
- state and action spaces
- closed form
- worst case
- multi agent
- multi armed bandit
- temporal difference
- minimax regret
- markov decision problems
- policy search
- evaluation function
- stationary policies
- continuous state and action spaces
- rl algorithms
- control problems
- partially observable
- continuous state spaces
- continuous state
- finite horizon
- machine learning
- optimal solution
- markov decision process
- average cost
- model based reinforcement learning