A Minimum Variance Path Principle for Accurate and Stable Score-Based Density Ratio Estimation

Wei Chen, Jiacheng Li, Shigui Li, Zhiqi Lin, Junmei Yang, John Paisley, Delu Zeng · 2026

Read A Minimum Variance Path Principle for Accurate and Stable Score-Based Density Ratio Estimation by Wei Chen, Jiacheng Li, Shigui Li, Zhiqi Lin, Junmei Yang, John Paisley, Delu Zeng (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-2602-00834.txt.

Description

Wei Chen and colleagues resolve a score-based paradox: theoretically path-independent, yet path-dependent in practice. They prove training goals differ from the ideal objective by a missed term, the score function's path variance. Their Minimum Variance Path principle cuts this variance, with a closed-form expression easing tuning. A Kumaraswamy Mixture Model shapes the path, learning data-tuned, low-variance paths without manual choice. Tuning the full objective gives more accurate and stable estimators with top results on challenging benchmarks.

Catalog record

CREATORWei Chen, Jiacheng Li, Shigui Li, Zhiqi Lin, Junmei Yang, John Paisley, Delu Zeng
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE114238
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00834

Browse the full library · About the Tower of Babel