First fully single-policy sample complexity bound for average-reward offline RL via bias span and policy hitting radius in weakly communicating MDPs.
For each s, a, there exists a setU4 sa ⊆ R with |U4 sa| ≤ S ntot 1−γ and random vectors(X4 u)u∈U 4sa such that 1) for allu ∈ U4 sa, X4 u is independent from S1 sa
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Optimal Single-Policy Sample Complexity and Transient Coverage for Average-Reward Offline RL
First fully single-policy sample complexity bound for average-reward offline RL via bias span and policy hitting radius in weakly communicating MDPs.