2504.01318

#1 Tail Bounds for Canonical $U$-Statistics and $U$-Processes with Unbounded Kernels [PDF] [Copy] [Kimi] [REL]

Authors: Abhishek Chakrabortty, Arun K. Kuchibhotla

In this paper, we prove exponential tail bounds for canonical (or degenerate) $U$-statistics and $U$-processes under exponential-type tail assumptions on the kernels. Most of the existing results in the relevant literature often assume bounded kernels or obtain sub-optimal tail behavior under unbounded kernels. We obtain sharp rates and optimal tail behavior under sub-Weibull kernel functions. Some examples from nonparametric and semiparametric statistics literature are considered.

Subjects: Statistics Theory , Probability

Publish: 2025-04-02 03:05:28 UTC

---