Time-uniform concentration bounds for iterative algorithms

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Pham, Tuan, Rinaldo, Alessandro, Sarkar, Purnamrita
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917099896569856
author Pham, Tuan
Rinaldo, Alessandro
Sarkar, Purnamrita
author_facet Pham, Tuan
Rinaldo, Alessandro
Sarkar, Purnamrita
contents We develop a new framework for deriving time-uniform concentration bounds for the output of stochastic sequential algorithms satisfying certain recursive inequalities akin to those defining the almost-supermartingale processes introduced by \cite{robbins1971convergence}. Our approach is of wide applicability, and can be deployed in settings in which exponential supermartingale processes, required by prevailing methodologies for anytime-valid concentration inequalities, are not readily available. Our results can be viewed as quantitative versions of the classical Robbins-Siegmund Lemma. We demonstrate the effectiveness of our method by providing new and optimal time-uniform concentration bounds for Oja's algorithm for streaming PCA, stochastic gradient descent, and stochastic approximations.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18273
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Time-uniform concentration bounds for iterative algorithms
Pham, Tuan
Rinaldo, Alessandro
Sarkar, Purnamrita
Statistics Theory
We develop a new framework for deriving time-uniform concentration bounds for the output of stochastic sequential algorithms satisfying certain recursive inequalities akin to those defining the almost-supermartingale processes introduced by \cite{robbins1971convergence}. Our approach is of wide applicability, and can be deployed in settings in which exponential supermartingale processes, required by prevailing methodologies for anytime-valid concentration inequalities, are not readily available. Our results can be viewed as quantitative versions of the classical Robbins-Siegmund Lemma. We demonstrate the effectiveness of our method by providing new and optimal time-uniform concentration bounds for Oja's algorithm for streaming PCA, stochastic gradient descent, and stochastic approximations.
title Time-uniform concentration bounds for iterative algorithms
topic Statistics Theory
url https://arxiv.org/abs/2511.18273