Assessing high-order effects in feature importance via predictability decomposition

Marlis Ontivero-Ortega 1 , Luca Faes 2 , Jesus M Cortes 3 , Daniele Marinazzo 4 , and Sebastiano Stramaglia 1 · 2024

Read Assessing high-order effects in feature importance via predictability decomposition by Marlis Ontivero-Ortega 1 , Luca Faes 2 , Jesus M Cortes 3 , Daniele Marinazzo 4 , and Sebastiano Stramaglia 1 (2024) 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-2412-09964.txt.

Description

Ontivero-Ortega and four co-authors split feature weight into solo and group parts in this twenty twenty-four paper. Feature weight scores help explain why a model makes its calls, but a feature may matter only when paired with others. The team builds on a known score that drops one feature at a time and checks the error change.

Inside, their method hunts for the feature group that makes the score as large as it can get. It also hunts for the group that makes the score as small as it can get. That split shows which partners add shared weight and which add fresh, joint weight. The team shows the method on test data that sorts two types of tiny particles apart.

Catalog record

CREATORMarlis Ontivero-Ortega 1 , Luca Faes 2 , Jesus M Cortes 3 , Daniele Marinazzo 4 , and Sebastiano Stramaglia 1
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-09964

Browse the full library · About the Tower of Babel