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. 2311.13010
29
2

Beyond Catoni: Sharper Rates for Heavy-Tailed and Robust Mean Estimation

21 November 2023
Shivam Gupta
Samuel B. Hopkins
Eric Price
ArXivPDFHTML
Abstract

We study the fundamental problem of estimating the mean of a ddd-dimensional distribution with covariance Σ≼σ2Id\Sigma \preccurlyeq \sigma^2 I_dΣ≼σ2Id​ given nnn samples. When d=1d = 1d=1, \cite{catoni} showed an estimator with error (1+o(1))⋅σ2log⁡1δn(1+o(1)) \cdot \sigma \sqrt{\frac{2 \log \frac{1}{\delta}}{n}}(1+o(1))⋅σn2logδ1​​​, with probability 1−δ1 - \delta1−δ, matching the Gaussian error rate. For d>1d>1d>1, a natural estimator outputs the center of the minimum enclosing ball of one-dimensional confidence intervals to achieve a 1−δ1-\delta1−δ confidence radius of 2dd+1⋅σ(dn+2log⁡1δn)\sqrt{\frac{2 d}{d+1}} \cdot \sigma \left(\sqrt{\frac{d}{n}} + \sqrt{\frac{2 \log \frac{1}{\delta}}{n}}\right)d+12d​​⋅σ(nd​​+n2logδ1​​​), incurring a 2dd+1\sqrt{\frac{2d}{d+1}}d+12d​​-factor loss over the Gaussian rate. When the dn\sqrt{\frac{d}{n}}nd​​ term dominates by a log⁡1δ\sqrt{\log \frac{1}{\delta}}logδ1​​ factor, \cite{lee2022optimal-highdim} showed an improved estimator matching the Gaussian rate. This raises a natural question: Is the 2dd+1\sqrt{\frac{2 d}{d+1}}d+12d​​ loss \emph{necessary} when the 2log⁡1δn\sqrt{\frac{2 \log \frac{1}{\delta}}{n}}n2logδ1​​​ term dominates? We show that the answer is \emph{no} -- we construct an estimator that improves over the above naive estimator by a constant factor. We also consider robust estimation, where an adversary is allowed to corrupt an ϵ\epsilonϵ-fraction of samples arbitrarily: in this case, we show that the above strategy of combining one-dimensional estimates and incurring the 2dd+1\sqrt{\frac{2d}{d+1}}d+12d​​-factor \emph{is} optimal in the infinite-sample limit.

View on arXiv
Comments on this paper