An Information-Theoretic Analysis of Thompson Sampling
Machine Learning
2015-06-09 v2
Abstract
We provide an information-theoretic analysis of Thompson sampling that applies across a broad range of online optimization problems in which a decision-maker must learn from partial feedback. This analysis inherits the simplicity and elegance of information theory and leads to regret bounds that scale with the entropy of the optimal-action distribution. This strengthens preexisting results and yields new insight into how information improves performance.
Cite
@article{arxiv.1403.5341,
title = {An Information-Theoretic Analysis of Thompson Sampling},
author = {Daniel Russo and Benjamin Van Roy},
journal= {arXiv preprint arXiv:1403.5341},
year = {2015}
}