Saved in:
Bibliographic Details
Main Authors: Arunachaleswaran, Eshwar Ram, Collina, Natalie, Roth, Aaron, Shi, Mirah
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2402.11410
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929529761562624
author Arunachaleswaran, Eshwar Ram
Collina, Natalie
Roth, Aaron
Shi, Mirah
author_facet Arunachaleswaran, Eshwar Ram
Collina, Natalie
Roth, Aaron
Shi, Mirah
contents Blasiok et al. [2023] proposed distance to calibration as a natural measure of calibration error that unlike expected calibration error (ECE) is continuous. Recently, Qiao and Zheng [2024] gave a non-constructive argument establishing the existence of an online predictor that can obtain $O(\sqrt{T})$ distance to calibration in the adversarial setting, which is known to be impossible for ECE. They leave as an open problem finding an explicit, efficient algorithm. We resolve this problem and give an extremely simple, efficient, deterministic algorithm that obtains distance to calibration error at most $2\sqrt{T}+1$.
format Preprint
id arxiv_https___arxiv_org_abs_2402_11410
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle An Elementary Predictor Obtaining $2\sqrt{T}+1$ Distance to Calibration
Arunachaleswaran, Eshwar Ram
Collina, Natalie
Roth, Aaron
Shi, Mirah
Machine Learning
Data Structures and Algorithms
Blasiok et al. [2023] proposed distance to calibration as a natural measure of calibration error that unlike expected calibration error (ECE) is continuous. Recently, Qiao and Zheng [2024] gave a non-constructive argument establishing the existence of an online predictor that can obtain $O(\sqrt{T})$ distance to calibration in the adversarial setting, which is known to be impossible for ECE. They leave as an open problem finding an explicit, efficient algorithm. We resolve this problem and give an extremely simple, efficient, deterministic algorithm that obtains distance to calibration error at most $2\sqrt{T}+1$.
title An Elementary Predictor Obtaining $2\sqrt{T}+1$ Distance to Calibration
topic Machine Learning
Data Structures and Algorithms
url https://arxiv.org/abs/2402.11410