On Stopping Times of Power-one Sequential Tests: Tight Lower and Upper Bounds

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Agrawal, Shubhada, Ramdas, Aaditya
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912350851825664
author Agrawal, Shubhada
Ramdas, Aaditya
author_facet Agrawal, Shubhada
Ramdas, Aaditya
contents We prove two lower bounds for stopping times of sequential tests between general composite nulls and alternatives. The first lower bound is for the setting where the type-1 error level $α$ approaches zero, and equals $\log(1/α)$ divided by a certain infimum KL divergence, termed $\operatorname{KL_{inf}}$. The second lower bound applies to the setting where $α$ is fixed and $\operatorname{KL_{inf}}$ approaches 0 (meaning that the null and alternative sets are not separated) and equals $c \operatorname{KL_{inf}}^{-1} \log \log \operatorname{KL_{inf}}^{-1}$ for a universal constant $c > 0$. We also provide a sufficient condition for matching the upper bounds and show that this condition is met in several special cases. Given past work, these upper and lower bounds are unsurprising in their form; our main contribution is the generality in which they hold, for example, not requiring reference measures or compactness of the classes.
format Preprint
id arxiv_https___arxiv_org_abs_2504_19952
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle On Stopping Times of Power-one Sequential Tests: Tight Lower and Upper Bounds
Agrawal, Shubhada
Ramdas, Aaditya
Statistics Theory
Machine Learning
We prove two lower bounds for stopping times of sequential tests between general composite nulls and alternatives. The first lower bound is for the setting where the type-1 error level $α$ approaches zero, and equals $\log(1/α)$ divided by a certain infimum KL divergence, termed $\operatorname{KL_{inf}}$. The second lower bound applies to the setting where $α$ is fixed and $\operatorname{KL_{inf}}$ approaches 0 (meaning that the null and alternative sets are not separated) and equals $c \operatorname{KL_{inf}}^{-1} \log \log \operatorname{KL_{inf}}^{-1}$ for a universal constant $c > 0$. We also provide a sufficient condition for matching the upper bounds and show that this condition is met in several special cases. Given past work, these upper and lower bounds are unsurprising in their form; our main contribution is the generality in which they hold, for example, not requiring reference measures or compactness of the classes.
title On Stopping Times of Power-one Sequential Tests: Tight Lower and Upper Bounds
topic Statistics Theory
Machine Learning
url https://arxiv.org/abs/2504.19952