Reward-Free RL is No Harder Than Reward-Aware RL in Linear Markov Decision Processes.
Andrew WagenmakerYifang ChenMax SimchowitzSimon S. DuKevin JamiesonPublished in: CoRR (2022)
Keyphrases
- markov decision processes
- reinforcement learning
- average reward
- reward function
- discounted reward
- total reward
- optimal policy
- policy iteration
- expected reward
- reinforcement learning algorithms
- state space
- finite state
- action space
- state and action spaces
- partially observable
- stationary policies
- dynamic programming
- model free
- function approximation
- rl algorithms
- model based reinforcement learning
- average cost
- machine learning
- transition matrices
- factored mdps
- infinite horizon
- stochastic games
- markov decision process
- state action
- temporal difference
- decision theoretic planning
- actor critic
- function approximators
- hierarchical reinforcement learning
- learning agent
- action sets
- reachability analysis
- planning under uncertainty
- finite horizon
- policy gradient
- semi markov decision processes
- action selection