A KL-LUCB Bandit Algorithm for Large-Scale Crowdsourcing

TL;DR

This paper introduces a new bandit algorithm combining lil-UCB and KL-LUCB for efficient best-arm identification in large-scale crowdsourcing, supported by theoretical bounds and experimental validation.

Contribution

It presents a novel anytime confidence bound for bounded distributions and fuses existing algorithms to improve large-scale crowdsourcing performance.

Findings

Proves a new confidence bound for bounded rewards

Demonstrates improved identification efficiency in experiments

Validates theoretical results with real-world data

Abstract

This paper focuses on best-arm identification in multi-armed bandits with bounded rewards. We develop an algorithm that is a fusion of lil-UCB and KL-LUCB, offering the best qualities of the two algorithms in one method. This is achieved by proving a novel anytime confidence bound for the mean of bounded distributions, which is the analogue of the LIL-type bounds recently developed for sub-Gaussian distributions. We corroborate our theoretical results with numerical experiments based on the New Yorker Cartoon Caption Contest.