Identifying Outlier Arms in Multi-Armed Bandit

We study a novel problem lying at the intersection of two areas: multi-armed banditand outlier detection. Multi-armed bandit is a useful tool to model the processof incrementally collecting data for multiple objects in a decision space. Outlierdetection is a powerful method to narrow down the attention to a few objects afterthe data for them are collected. However, no one has studied how to detect outlierobjects while incrementally collecting data for them, which is necessary when datacollection is expensive. We formalize this problem as identifying outlier arms in amulti-armed bandit. We propose two sampling strategies with theoretical guarantee,and analyze their sampling efficiency. Our experimental results on both syntheticand real data show that our solution saves 70-99% of data collection cost frombaseline while having nearly perfect accuracy.