On Unified and Sharpened CMI Bounds for Generalization Errors

Yang Lu, Matthias Frey, Margreta Kuijper, and Jingge Zhu · 2026

Read On Unified and Sharpened CMI Bounds for Generalization Errors by Yang Lu, Matthias Frey, Margreta Kuijper, and Jingge Zhu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-21056.txt.

Description

Yang Lu and colleagues present a unified family of conditional mutual information bounds for generalization error, built on leave-m-out cross-validation error. The unified bound envelops and reproduces many known conditional mutual information bounds and bridges mutual information and conditional mutual information bounds as m grows large. It also yields new, sharper bounds. When loss is bounded, trimming conditional terms tightens the bounds further, and examples show gains over prior results.

Catalog record

CREATORYang Lu, Matthias Frey, Margreta Kuijper, and Jingge Zhu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE186340
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.21056

Browse the full library · About the Tower of Babel