Efficient Counterfactual Learning from Bandit Feedback
By Yusuke Narita, Shota Yasui, Kohei Yata. 2018
49 citations recorded by the source corpus.
What is the most statistically efficient way to do off-policy optimization with batch data from bandit feedback? For log data generated by contextual bandit algorithms, we consider offline estimators for the expected reward from a counterfactual policy. Our estimators are shown to have lowest variance in a wide class of estimators, achieving variance reduction relative to standard estimators. We then apply our estimators to improve advertisement design by a major advertisement company. Consistent with the theoretical result, our estimators allow us to improve on the existing bandit algorithm with more statistical confidence compared to a state-of-theart benchmark.
Subfields: Artificial Intelligence, Electrical and Electronic Engineering, Management Science and Operations Research
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Scott Niekum
- Masatoshi Uehara
- Nathan Kallus
- Josiah P. Hanna
- Peter Stone
- Sergey Levine
- Ofir Nachum
- Nan Jiang
In Electrical and Electronic Engineering
In Management Science and Operations Research
- Masatoshi Uehara
- Nathan Kallus
- Scott Niekum
- Peter Stone
- Josiah P. Hanna
- Andrew Bennett
- Susan Athey
- Stefan Wager
Citations in the corpus
- Intrinsically Efficient, Stable, and Bounded Off-Policy Evaluation for Reinforcement Learning (2019, background): “When this model is correct, β̂dm is both consistent for β∗ and locally efficient in that its asymptotic MSE is minimized among the class of all estimators consistent for β∗ [19, 33].”
- Importance Sampling Policy Evaluation with an Estimated Behavior Policy (2018, background): “However, research in the multi-armed bandit (Li et al., 2015; Narita et al., 2019), causal inference (Hirano et al.”