Enhancing Visual Feature Attribution via Weighted Integrated Gradients

Tran Duc Tuan Kien, Nguyen Trong Tam, Nguyen Hoang Son, Khoat Than, Duc Anh Nguyen · 2025

Read Enhancing Visual Feature Attribution via Weighted Integrated Gradients by Tran Duc Tuan Kien, Nguyen Trong Tam, Nguyen Hoang Son, Khoat Than, Duc Anh Nguyen (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-2505-03201.txt.

Description

The authors weigh base images to firm up visual feature attribution in a 2025 paper. Integrated Gradients grades which pixels drove an image call, but its marks swing with the base pick. Past blends treat all bases as peers, which adds noise. Their Weighted form scores each base for fit, then weights by that score per input. Core rules of the stock method hold in the new blend.

On set image banks, the method beat the flat blend by up to 36 percent. Gains held on both classic and transformer nets. Extra score checks form the cost of that gain. The authors frame it as a truth-for-time trade, not a speed play.

Catalog record

CREATORTran Duc Tuan Kien, Nguyen Trong Tam, Nguyen Hoang Son, Khoat Than, Duc Anh Nguyen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-03201

Browse the full library · About the Tower of Babel