When is Agnostic Reinforcement Learning Statistically Tractable?
Zeyu JiaGene LiAlexander RakhlinAyush SekhariNati SrebroPublished in: NeurIPS (2023)
Keyphrases
- reinforcement learning
- function approximation
- np complete
- reinforcement learning algorithms
- direct policy search
- action selection
- model free
- transfer learning
- state space
- computational complexity
- multi agent
- machine learning
- np hard
- stochastic approximation
- relational reinforcement learning
- temporal difference learning
- learning algorithm
- optimal control
- temporal difference
- robotic control
- data sets
- computational problems
- control problems
- optimal policy
- case study
- computationally tractable
- markov decision processes
- real world
- real time