arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2302.05686math.STcs.LGstat.MLstat.TH
收藏

二阶U-统计量的高维收敛定理及其在基于核的检验中的应用

A High-dimensional Convergence Theorem for U-statistics with Applications to Kernel-based Testing

University College London(伦敦大学学院);Imperial College London(帝国理工学院);Alan Turing Institute(艾伦·图灵研究所)

Kevin H. Huang, Xing Liu, Andrew B. Duncan, Axel Gandy

AI总结:

证明了二阶U-统计量在高维下的收敛定理,发现其极限分布经历相变,并应用于MMD和KSD两种核检验,正确预测了高维下检验功效与维度及带宽的关系。

摘要:

我们证明了二阶U-统计量的收敛定理,其中数据维度$d$允许随样本量$n$变化。我们发现,U-统计量的极限分布会经历从非退化高斯极限到退化极限的相变,这仅取决于一个矩比,而与其退化性无关。一个令人惊讶的结论是,高维下的非退化U-统计量可能具有非高斯极限,且方差更大、分布不对称。我们的界对任意有限的$n$和$d$均成立,不依赖于底层函数的各个特征值,并在温和假设下与维度无关。作为应用,我们将理论应用于两种流行的基于核的分布检验MMD和KSD,其高维性能一直难以研究。在简单的经验设置中,我们的结果正确预测了固定阈值下检验功效如何随$d$和带宽变化。

英文摘要:

We prove a convergence theorem for U-statistics of degree two, where the data dimension $d$ is allowed to scale with sample size $n$. We find that the limiting distribution of a U-statistic undergoes a phase transition from the non-degenerate Gaussian limit to the degenerate limit, regardless of its degeneracy and depending only on a moment ratio. A surprising consequence is that a non-degenerate U-statistic in high dimensions can have a non-Gaussian limit with a larger variance and asymmetric distribution. Our bounds are valid for any finite $n$ and $d$, independent of individual eigenvalues of the underlying function, and dimension-independent under a mild assumption. As an application, we apply our theory to two popular kernel-based distribution tests, MMD and KSD, whose high-dimensional performance has been challenging to study. In a simple empirical setting, our results correctly predict how the test power at a fixed threshold scales with $d$ and the bandwidth.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑