Near Instance-Optimal PAC Reinforcement Learning for Deterministic MDPs.
Andrea TirinzoniAymen Al MarjaniEmilie KaufmannPublished in: NeurIPS (2022)
Keyphrases
- reinforcement learning
- markov decision processes
- dynamic programming
- state space
- action sets
- optimal policy
- optimal control
- average cost
- function approximation
- markov decision process
- state and action spaces
- initially unknown
- approximate dynamic programming
- average reward
- finite horizon
- model free
- stationary policies
- reinforcement learning algorithms
- continuous state spaces
- policy search
- deterministic domains
- policy iteration
- control policies
- control problems
- factored mdps
- policy evaluation
- upper bound
- temporal difference
- markov decision problems
- pac learning
- learning algorithm