Derives an order-explicit large deviation bound for high-dimensional U-statistics from their Hájek projections, yielding new concentration results and consistency for resampling-based confidence intervals around subsampled kernel regression estimators.
Detailed proof of Nazarov's inequality
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
abstract
The purpose of this note is to provide a detailed proof of Nazarov's inequality stated in Lemma A.1 in Chernozhukov, Chetverikov, and Kato (2017, Annals of Probability).
verdicts
UNVERDICTED 3representative citing papers
Derived rates of order up to n^{-1/6} log^4(n S A) for the high-dimensional CLT of averaged asynchronous Q-learning iterates, plus a general martingale-difference CLT.
Sparsity of regression parameters or differential parameters is not necessary for consistent multiple change point detection in high-dimensional linear regression; a covariance discrepancy scan is statistically and computationally more efficient.
citing papers explorer
-
Order-Explicit Linearization of High-Dimensional $U$-Statistics
Derives an order-explicit large deviation bound for high-dimensional U-statistics from their Hájek projections, yielding new concentration results and consistency for resampling-based confidence intervals around subsampled kernel regression estimators.
-
Gaussian Approximation for Asynchronous Q-learning
Derived rates of order up to n^{-1/6} log^4(n S A) for the high-dimensional CLT of averaged asynchronous Q-learning iterates, plus a general martingale-difference CLT.
-
Detection and inference of changes in high-dimensional linear regression with non-sparse structures
Sparsity of regression parameters or differential parameters is not necessary for consistent multiple change point detection in high-dimensional linear regression; a covariance discrepancy scan is statistically and computationally more efficient.