Time-Uniform Confidence Spheres for Means of Random Vectors

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Chugg, Ben, Wang, Hongjian, Ramdas, Aaditya
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913837338329088
author Chugg, Ben
Wang, Hongjian
Ramdas, Aaditya
author_facet Chugg, Ben
Wang, Hongjian
Ramdas, Aaditya
contents We study sequential mean estimation in $\mathbb{R}^d$. In particular, we derive time-uniform confidence spheres -- confidence sphere sequences (CSSs) -- which contain the mean of random vectors with high probability simultaneously across all sample sizes. Our results include a dimension-free CSS for log-concave random vectors, a dimension-free CSS for sub-Gaussian random vectors, and CSSs for sub-$ψ$ random vectors (which includes sub-gamma, sub-Poisson, and sub-exponential distributions). Many of our results are optimal. For sub-Gaussian distributions we also provide a CSS which tracks a time-varying mean, generalizing Robbins' mixture approach to the multivariate setting. Finally, we provide several CSSs for heavy-tailed random vectors (two moments only). Our bounds hold under a martingale assumption on the mean and do not require that the observations be iid. Our work is based on PAC-Bayesian theory and inspired by an approach of Catoni and Giulini.
format Preprint
id arxiv_https___arxiv_org_abs_2311_08168
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Time-Uniform Confidence Spheres for Means of Random Vectors
Chugg, Ben
Wang, Hongjian
Ramdas, Aaditya
Statistics Theory
Information Theory
Methodology
Machine Learning
We study sequential mean estimation in $\mathbb{R}^d$. In particular, we derive time-uniform confidence spheres -- confidence sphere sequences (CSSs) -- which contain the mean of random vectors with high probability simultaneously across all sample sizes. Our results include a dimension-free CSS for log-concave random vectors, a dimension-free CSS for sub-Gaussian random vectors, and CSSs for sub-$ψ$ random vectors (which includes sub-gamma, sub-Poisson, and sub-exponential distributions). Many of our results are optimal. For sub-Gaussian distributions we also provide a CSS which tracks a time-varying mean, generalizing Robbins' mixture approach to the multivariate setting. Finally, we provide several CSSs for heavy-tailed random vectors (two moments only). Our bounds hold under a martingale assumption on the mean and do not require that the observations be iid. Our work is based on PAC-Bayesian theory and inspired by an approach of Catoni and Giulini.
title Time-Uniform Confidence Spheres for Means of Random Vectors
topic Statistics Theory
Information Theory
Methodology
Machine Learning
url https://arxiv.org/abs/2311.08168