A framework for baseline-corrected values in EFGs is introduced, with new baselines including a predictive one that is provably optimal for zero-variance estimates under certain sampling schemes.
Doubly robust off-policy value evaluation for reinforcement learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.GT 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Low-Variance and Zero-Variance Baselines for Extensive-Form Games
A framework for baseline-corrected values in EFGs is introduced, with new baselines including a predictive one that is provably optimal for zero-variance estimates under certain sampling schemes.