30
9

Design based incomplete U-statistics

Abstract

U-statistics are widely used in fields such as economics, machine learning, and statistics. However, while they enjoy desirable statistical properties, they have an obvious drawback in that the computation becomes impractical as the data size nn increases. Specifically, the number of combinations, say mm, that a U-statistic of order dd has to evaluate is O(nd)O(n^d). Many efforts have been made to approximate the original U-statistic using a small subset of combinations since Blom (1976), who referred to such an approximation as an incomplete U-statistic. To the best of our knowledge, all existing methods require mm to grow at least faster than nn, albeit more slowly than ndn^d, in order for the corresponding incomplete U-statistic to be asymptotically efficient in terms of the mean squared error. In this paper, we introduce a new type of incomplete U-statistic that can be asymptotically efficient, even when mm grows more slowly than nn. In some cases, mm is only required to grow faster than n\sqrt{n}. Our theoretical and empirical results both show significant improvements in the statistical efficiency of the new incomplete U-statistic.

View on arXiv
Comments on this paper