Data-driven Error Estimation: Excess Risk Bounds without Class Complexity as Input

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Krishnamurthy, Sanath Kumar, Lyubarskaja, Anna, Brunskill, Emma, Athey, Susan
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917246038704128
author Krishnamurthy, Sanath Kumar
Lyubarskaja, Anna
Brunskill, Emma
Athey, Susan
author_facet Krishnamurthy, Sanath Kumar
Lyubarskaja, Anna
Brunskill, Emma
Athey, Susan
contents Constructing confidence intervals that are simultaneously valid across a class of estimates is central to tasks such as multiple mean estimation, generalization guarantees, and adaptive experimental design. We frame this as an ``error estimation problem," where the goal is to determine a high-probability upper bound on the maximum error for a class of estimates. We propose an entirely data-driven approach that derives such bounds for both finite and infinite class settings, naturally adapting to a potentially unknown correlation structure of random errors. Notably, our method does not require class complexity as an input, overcoming a major limitation of existing approaches. We present our simple yet general solution and demonstrate applications to simultaneous confidence intervals, excess-risk control and optimizing exploration in contextual bandit algorithms.
format Preprint
id arxiv_https___arxiv_org_abs_2405_04636
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Data-driven Error Estimation: Excess Risk Bounds without Class Complexity as Input
Krishnamurthy, Sanath Kumar
Lyubarskaja, Anna
Brunskill, Emma
Athey, Susan
Machine Learning
62G15, 68Q32, 62G05, 62L05,
G.3
Constructing confidence intervals that are simultaneously valid across a class of estimates is central to tasks such as multiple mean estimation, generalization guarantees, and adaptive experimental design. We frame this as an ``error estimation problem," where the goal is to determine a high-probability upper bound on the maximum error for a class of estimates. We propose an entirely data-driven approach that derives such bounds for both finite and infinite class settings, naturally adapting to a potentially unknown correlation structure of random errors. Notably, our method does not require class complexity as an input, overcoming a major limitation of existing approaches. We present our simple yet general solution and demonstrate applications to simultaneous confidence intervals, excess-risk control and optimizing exploration in contextual bandit algorithms.
title Data-driven Error Estimation: Excess Risk Bounds without Class Complexity as Input
topic Machine Learning
62G15, 68Q32, 62G05, 62L05,
G.3
url https://arxiv.org/abs/2405.04636