Entropy Regularization for Population Estimation

Entropy regularization is known to improve exploration in sequential decision-making problems. We show that this same mechanism can also lead nearly unbiased and lower-variance estimates of the mean reward optimize-and-estimate structured bandit setting. Mean estimation (i.e., population estimation) tasks have recently been shown be essential for public policy settings where legal constraints often require precise metrics. leveraging entropy KL divergence yield a better trade-off between estimator variance than existing baselines, all while remaining unbiased. These properties illustrate an exciting potential bringing together optimal literature.