ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2410.19244
11
0

Universality of Estimator for High-Dimensional Linear Models with Block Dependency

25 October 2024
Toshiki Tsuda
Masaaki Imaizumi
ArXivPDFHTML
Abstract

We study the universality property of estimators for high-dimensional linear models, which exhibits a distribution of estimators is independent of whether covariates follows a Gaussian distribution. Recent high-dimensional statistics require covariates to strictly follow a Gaussian distribution to reveal precise properties of estimators. To relax the Gaussianity requirement, the existing literature has studied the conditions that allow estimators to achieve universality. In particular, independence of each element of the high-dimensional covariates plays an important role. In this study, we focus on high-dimensional linear models and covariates with block dependencies, where elements of covariates only within a block can be dependent, then show that estimators for the model maintain the universality. Specifically, we prove that a distribution of estimators with Gaussian covariates is approximated by an estimator with non-Gaussian covariates with same moments, with the setup of block dependence. To establish the result, we develop a generalized Lindeberg principle to handle block dependencies and derive new error bounds for correlated elements of covariates. We also apply our result of the universality to a distribution of robust estimators.

View on arXiv
Comments on this paper