Structured Prediction with Abstention via the Lovász Hinge

Jessie Finocchiaro [email protected], Rafael Frongillo [email protected], Enrique Nueve [email protected] · 2025

Read Structured Prediction with Abstention via the Lovász Hinge by Jessie Finocchiaro [email protected], Rafael Frongillo [email protected], Enrique Nueve [email protected] (2025) 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-2505-06446.txt.

Description

Jessie Finocchiaro, Rafael Frongillo, and Enrique Nueve study a loss function used for joint binary predictions, in a 2025 paper. The Lovasz hinge is popular in tasks such as image segmentation, where several linked yes-no predictions are scored together by a set function. Whether that loss is consistent, meaning it truly targets the desired problem, was an open question.

The authors show it is inconsistent with its intended target unless the scoring function is of a simple additive kind. They then find the problem it does target: a form of structured prediction that may abstain on any subset of the predictions. They build link functions that work for a broad class of scoring functions and give conditions for a tight match. Tests show the abstention view aids interpretability. A multi-class version is also built from the binary case.

Catalog record

CREATORJessie Finocchiaro [email protected], Rafael Frongillo [email protected], Enrique Nueve [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE135 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-06446

Browse the full library · About the Tower of Babel