Near-Optimal Confidence Sequences for Bounded Random Variables

TL;DR

This paper introduces a near-optimal confidence sequence for bounded random variables, improving the accuracy and efficiency of online inference in sequential decision-making and adaptive sampling tasks.

Contribution

It develops a new confidence sequence based on Bentkus' concentration results, surpassing traditional methods like Hoeffding and Bernstein inequalities in performance.

Findings

Outperforms existing confidence sequences in synthetic coverage tests.

Effective in adaptive stopping algorithms.

Provides tighter bounds for bounded random variables.

Abstract

Many inference problems, such as sequential decision problems like A/B testing, adaptive sampling schemes like bandit selection, are often online in nature. The fundamental problem for online inference is to provide a sequence of confidence intervals that are valid uniformly over the growing-into-infinity sample sizes. To address this question, we provide a near-optimal confidence sequence for bounded random variables by utilizing Bentkus' concentration results. We show that it improves on the existing approaches that use the Cram{\'e}r-Chernoff technique such as the Hoeffding, Bernstein, and Bennett inequalities. The resulting confidence sequence is confirmed to be favorable in both synthetic coverage problems and an application to adaptive stopping algorithms.